DE102017213160A1 - Compilation for node device GPU-based parallel processing - Google Patents

Abstract

A device may include a processor and memory to store instructions that cause the processor to perform operations, including: in response to a determination that a GPU is available to a node device, determining whether a task routine can be compiled to a GPU Create task routine for execution by the GPU to cause execution of multiple instances of a task routine task at least partially in parallel without dependencies therebetween; and in response to a determination that the task routine can be compiled to generate the GPU task routine: using a conversion rule to transform the task routine into the GPU task routine; Compiling the GPU task routine for execution by the GPU; and assigning the task with a record partition to the node device to allow the GPU to perform the multiple instances with the record partition.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of priority according to 35 U.S.C. & 119 (e) from US provisional patent application serial no. No. 62 / 289,484 filed on Feb. 1, 2016, US Provisional Patent Application Ser. No. 62 / 379,512, filed on August 25, 2016, and US provisional patent application serial no. 62 / 394,411, filed September 14, 2016, the entirety of which is incorporated herein by reference. This application is a continuation of and claims the benefit of priority according to 35 U.S.C. §120 from the US patent application serial no. 15 / 422,154, filed concurrently herewith on Feb. 1, 2017, the entirety of which is incorporated herein by reference.

BACKGROUND

It has become increasingly common to use lattices from numerous node devices to perform analyzes of large data sets (eg, commonly referred to as "big data") in a distributed fashion in which tasks of analysis routines are at least partially parallel over a plurality of selected ones of the node devices be performed. In the operation of such grids, there are often competing goals in assigning the tasks of analysis routines to the node devices, including the goal of using the processing resources of each node device as fully and uninterrupted as possible, and the destination, the processing and storage resources of each node device, and the network bandwidth resources of the node device To use grids more effectively. What is added to the complexity of addressing such competing targets is the common practice of sharing lattice resources by causing the tasks of multiple analysis routines to be concurrently performed by the node devices of the lattice. A classic approach to assigning tasks is simply to assign the next task to be performed to the node device that is closest to available processing resources. Unfortunately, this classic approach is a decision to allow significant inefficiencies in the use of memory and network bandwidths.

SHORT VERSION

This summary is not intended to identify only key or essential features of the described subject matter, nor should it be used in isolation to determine the scope of protection of the described subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification of this patent, any or all of the drawings, and any claims.

A device may include a processor and memory for storing instructions that, when executed by the processor, cause the processor to derive an assignment of performing a first task with a first set of data for each node device of a plurality of node devices the first record is divisible into a plurality of partitions; a first node device is assigned to the plurality of node devices to perform the first task with a first partition of the plurality of partitions of the first data set; and a second node device is assigned to the plurality of node devices to perform the first task with a second partition of the plurality of partitions of the first data set. The processor may be caused to transmit an indication of the assignment of performing the first task with the first partition to the first node device to cause the first task to perform the first task by the first node device and to store at least a portion of the first task cause first partition within the volatile memory of the first node device; Transmitting an indication of the assignment of performing the first task with the second partition to the second node device to cause the second task to perform the first task with the second partition, and storing at least a portion of the second partition within the volatile memory to cause the second node device; Receiving, by the first node device, an indication from the first node device of completion of performing the first task with the first partition, such that the first node device is available to assign the performance to another task; Delaying the assignment of the performance of a second task on the second partition to the first node device upon receiving the indication of the completion of the performance of the first task with the first partition by the first node device for a predetermined period of time, despite the second task being performed on the second partition second partition is to be performed, and despite availability of the first node as a result of completion of the execution of the first task with the first partition; and determining whether an indication of the completion of performing the first task with the second partition by the second node device so that the second node device is available to assign the performance of another task is received by the second node device within the predetermined time period. In response to receiving the indication of completion of the first task with the second partition by the second node device within the predetermined time period, the processor may be caused to allocate the second task on the second partition to the second node device for accessing at least one of the second task Allow part of the second partition within the volatile memory of the second node device; and transmitting to the second node device an indication of the assignment of performing the second task on the second partition to avoid fetching the second partition by the first node device. In response to a failure of the second task to terminate the first task with the second partition by the second node device within the predetermined period of time, the processor may be caused to allocate the second task on the second partition to the first node device; and transmitting to the first node device an indication of the assignment of performing the second task on the second partition to cause the second partition to fetch the second partition.

The processor may be caused to perform operations including deriving the predetermined period of time from at least one measurement of a time period between the transmission of an assignment of performing the first task to a node device of the plurality of node devices and receiving an indication of completion of performing the first task of the node device.

The processor may be caused to perform operations, including: determining a number of node devices of the plurality of node devices that are available to perform the first task; and deriving a division of the first record into the plurality of partitions of the first record based on the number of node devices and metadata descriptive of a manner in which the first record is organized. The first record may be stored within one or more storage devices; the processor may be caused to perform operations, including retrieving the metadata from the one or more storage devices; transmitting the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to retrieve the first partition from the one or more memory devices; and transmitting the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to retrieve the second partition from the one or more storage devices.

The apparatus may include at least one volatile memory component coupled to the processor and the processor may be caused to perform operations, including assigning to the processor the performance of the first task with a third partition of the plurality of partitions of the first data set; Storing at least a portion of the third partition within the at least one volatile memory component; and performing the first task with the third partition.

The processor may be caused to perform operations, including, for each node device of a subset of the plurality of node devices, deriving an assignment to retrieve and store one of the plurality of partitions of the first data set from one or more memory devices to determine the use of each node device Enabling subset as a backup node device to respond to a failure of one of the node devices of the plurality of node devices, wherein: a third node device is assigned to the plurality of node devices to perform the first task with a third partition of the plurality of partitions of the first data set ; and the third node is assigned to retrieve and store the second partition from the one or more storage devices to enable use of the third node device as a backup node device to respond to a failure of the second node device. The processor may be caused to receive an indication from the third node device of the termination of performing the first task with the third partition by the third node device during the predetermined time period such that the third node device is available to assign the performance to another task. In response to receiving the indication of completion of the first task with the third partition by the third node device during the predetermined time period and in response to failure to receive the indication of completion of the first task with the second partition by the second node device within the predetermined time period For example, the processor may be caused to assign execution of the second task on the second Partition to the third node device; and transmitting to the third node device an indication of the assignment of performing the second task on the second partition.

The implementations of the first task with the first and second partitions may include using the first and second partitions as inputs to implement the first task to create corresponding partitions of a second data set; and performing the second task on the second partition may include using the second partition as an input to perform the second task to generate a corresponding partition of a third data set. The transmission of the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to: retrieve the first partition from one or more storage devices; Using at least a portion of the first partition stored within the volatile memory of the first node device as an input for performing the first task by the first node device; and transmitting the indication of completion of performing the first task with the first partition while leaving at least a portion of the first partition stored within the volatile memory of the first node device. The transmission of the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to: retrieve the second partition from one or more storage devices; Using at least a portion of the second partition stored within the volatile memory of the second node device as an input for performing the first task by the second node device; and transmitting the indication of completion of performing the first task with the second partition while leaving at least a portion of the second partition stored within the volatile memory of the second node device. The transmission of the indication of assignment of the second job on the second partition to the second node device may cause the second node device to use at least a portion of the second partition remaining within the volatile memory of the second node device as an input to perform the second second task is stored by the second node device to minimize accesses to the second partition stored within the non-volatile memory of the second node device.

The implementations of the first task with the first and second partitions may include implementations of the first task to generate the first and second partitions as outputs of the first task using corresponding partitions of a second data set as inputs; and performing the second task on the second partition may include using the second partition as an input to perform the second task to generate a corresponding partition of a third data set. The transmission of the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to: generate the first partition as an output of performing the first task by the first node device; and transmitting the indication of completion of performing the first task with the first partition while leaving at least a portion of the first partition stored within the volatile memory of the first node device. The transmission of the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to: generate the second partition as an output of performing the first task by the second node device; and transmitting the indication of termination of performance of the first task with the second partition; while at least a portion of the second partition remains stored within the volatile memory of the second node device. The transmission of the indication of assignment of the second job on the second partition to the second node device may cause the second node device to use at least a portion of the second partition remaining within the volatile memory of the second node device as an input to perform the second second task is stored by the second node device to minimize accesses to the second partition stored within the non-volatile memory of the second node device.

A computer program product concretely embodied in a non-transitory machine-readable storage medium, the computer program product comprising instructions operable to cause a processor to perform operations including, for each node device of a plurality of node devices, deriving an assignment of the execution a first task having a first record, wherein: the first record is divisible into a plurality of partitions; a first node device is assigned to the plurality of node devices to perform the first task with a first partition of the plurality of partitions of the first data set; and a second node device is assigned to the plurality of node devices to perform the first task with a second partition of the plurality of partitions of the first data set. The processor can to perform operations, including: transmitting to the first node device an indication of the assignment of performing the first task with the first partition to cause the first task to perform the first task by the first node device and to store at least one of the first task Cause part of the first partition within the volatile memory of the first node device; Transmitting an indication of the assignment of performing the first task with the second partition to the second node device to cause the second task to perform the first task with the second partition, and storing at least a portion of the second partition within the volatile memory to cause the second node device; Receiving, by the first node device, an indication from the first node device of completion of performing the first task with the first partition, such that the first node device is available to assign the performance to another task; Delaying the assignment of the performance of a second task on the second partition to the first node device upon receiving the indication of the completion of the performance of the first task with the first partition by the first node device for a predetermined period of time, despite the second task being performed on the second partition second partition is to be performed, and despite availability of the first node as a result of the completion of the execution of the first task with the first partition; and determining whether an indication of the completion of performing the first task with the second partition by the second node device so that the second node device is available to assign the performance of another task is received by the second node device within the predetermined time period. In response to receiving the indication of completion of the first task with the second partition by the second node device within the predetermined time period, the processor may be caused to allocate the second task on the second partition to the second node device for accessing at least one of the second task Allow part of the second partition within the volatile memory of the second node device; and transmitting to the second node device an indication of the assignment of performing the second task on the second partition to avoid fetching the second partition by the first node device. In response to a failure of the second task to terminate the first task with the second partition by the second node device within the predetermined period of time, the processor may be caused to allocate the second task on the second partition to the first node device; and transmitting to the first node device an indication of the assignment of performing the second task on the second partition to cause the second partition to fetch the second partition.

The processor may be caused to perform operations including deriving the predetermined period of time from at least one measurement of a time period between the transmission of an assignment of performing the first task to a node device of the plurality of node devices and receiving an indication of completion of performing the first task of the node device.

The processor is caused to perform operations, including: determining a number of node devices of the plurality of node devices that are available to perform the first task; and deriving a division of the first record into the plurality of partitions of the first record based on the number of node devices and metadata descriptive of a manner in which the first record is organized. The first record may be stored within one or more storage devices; the processor may be caused to perform operations comprising retrieving the metadata from the one or more storage devices; transmitting the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to retrieve the first partition from the one or more memory devices; and transmitting the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to retrieve the second partition from the one or more storage devices.

The processor is caused to perform operations, including assigning to the processor the performance of the first task with a third partition of the plurality of partitions of the first set of data; Storing at least a portion of the third partition within at least one volatile memory component coupled to the processor; and performing the first task with the third partition.

The processor may be caused to perform operations including, for each node device of a subset of the plurality of node devices, deriving an assignment to retrieve one of the plurality of partitions of the first data set from one or more memory devices and store to allow the use of each node device of the subset as a backup node device to respond to a failure of one of the node devices of the plurality of node devices, wherein: a third node device is assigned to the plurality of node devices to perform the first task perform third partition of the plurality of partitions of the first record; and the third node is assigned to retrieve and store the second partition from the one or more storage devices to enable use of the third node device as a backup node device to respond to a failure of the second node device. The processor may be caused to receive an indication from the third node device of the termination of performing the first task with the third partition by the third node device during the predetermined time period such that the third node device is available to assign the performance to another task. The processor may, in response to receipt of the indication of completion of the first task with the third partition, by the third node device during the predetermined time period and in response to a lack of receipt of the indication of termination of the first task with the second partition by the second node device within the predetermined period of time, are arranged to: assign execution of the second task on the second partition to the third node device; and transmitting to the third node device an indication of the assignment of performing the second task on the second partition.

The implementations of the first task with the first and second partitions may include using the first and second partitions as inputs to implement the first task to create corresponding partitions of a second data set; and the Performing the second task on the second partition may include using the second partition as an input to perform the second task to generate a corresponding partition of a third data set. The transmission of the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to: retrieve the first partition from one or more storage devices; Using at least a portion of the first partition stored within the volatile memory of the first node device as an input for performing the first task by the first node device; and transmitting the indication of completion of performing the first task with the first partition while leaving at least a portion of the first partition stored within the volatile memory of the first node device. The transmission of the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to: retrieve the second partition from one or more storage devices; Using at least a portion of the second partition stored within the volatile memory of the second node device as an input for performing the first task by the second node device; and transmitting the indication of completion of performing the first task with the second partition while leaving at least a portion of the second partition stored within the volatile memory of the second node device. The transmission of the indication of assignment of the second job on the second partition to the second node device may cause the second node device to use at least a portion of the second partition remaining within the volatile memory of the second node device as an input to perform the second second task is stored by the second node device to minimize accesses to the second partition stored within the non-volatile memory of the second node device.

The implementations of the first task with the first and second partitions may include implementations of the first task to generate the first and second partitions as outputs of the first task using corresponding partitions of a second data set as inputs; and performing the second task on the second partition may include using the second partition as an input to perform the second task to generate a corresponding partition of a third data set. The transmission of the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to: generate the first partition as an output of performing the first task by the first node device; and transmitting the indication of completion of performing the first task with the first partition while leaving at least a portion of the first partition stored within the volatile memory of the first node device. The transmission of the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to: generate the second partition as an output of performing the first task by the second node device; and transmitting the indication of completion of performing the first task with the second partition while leaving at least a portion of the second partition stored within the volatile memory of the second node device. The transmission of the indication of assignment of the second job on the second partition to the second node device may cause the second node device to use at least a portion of the second partition remaining within the volatile memory of the second node device as an input to perform the second second task is stored by the second node device to minimize accesses to the second partition stored within the non-volatile memory of the second node device.

A computer-implemented method may, for each node device of a plurality of node devices, derive at a coordinator from an assignment of performing a first task to a first set, the first set being divisible into a plurality of partitions, and the deriving may include: deriving a first assignment of a first node device of the plurality of node devices to perform the first task with a first partition of the plurality of partitions of the first data set; and deriving a second assignment of a second node device of the plurality of node devices to perform the first task with a second partition of the plurality of partitions of the first data set. The method may include transmitting an indication of the assignment of performing the first task with the first partition to the first node device to cause the first task to perform the first task by the first node device and storing at least a portion of the first partition within the volatile memory of the first node device; Transmitting an indication of the assignment of performing the first task with the second partition to the second node device to cause the second task to perform the first task with the second partition, and storing at least a portion of the second partition within the volatile memory to cause the second node device; Receiving, at the coordinator, an indication from the first node device of the completion of the first task's execution of the first partition by the first node device, such that the first node device is available to assign the performance of another task; Delaying the assignment of the performance of a second task on the second partition to the first node device upon receiving the indication of the completion of the performance of the first task with the first partition by the first node device for a predetermined period of time, despite the second task being performed on the second partition second partition is to be performed, and despite availability of the first node as a result of the completion of the execution of the first task with the first partition; and determining, at the coordinating device, whether an indication of the completion of the performance of the first task with the second partition by the second node device so that the second node device is available to assign the performance of another task, from the second node device within the predetermined time period Will be received. The method may include, in response to receiving the indication of completion of the first task with the second partition by the second node device within the predetermined time period: assigning the second task on the second partition to the second node device to access at least a portion enable the second partition within the volatile memory of the second node device; and transmitting to the second node device an indication of the assignment of performing the second task on the second partition to avoid fetching the second partition by the first node device.

The method may include, at the coordinating device, deriving the predetermined period of time from at least one measurement of a time period between the transmission of an assignment of performing the first task to a node device of the plurality of node devices and receiving an indication of completion of the first task Include node device.

The method may include determining, at the coordinating device, a number of node devices of the plurality of node devices that are available to perform the first task; and deriving, at the coordinating device, a distribution of the first record into the plurality of partitions of the first record based on the number of node devices and metadata descriptive of a manner in which the first record is organized. The first record is stored within one or more storage devices; the method may include retrieving the metadata, by the coordinating device, from the one or more storage devices; transmitting the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to retrieve the first partition from the one or more memory devices; and transmitting the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to retrieve the second partition from the one or more storage devices.

The method may include assigning, to the coordinator, the performance of the first task with a third partition of the plurality of partitions of the first set of data; Storing at least a portion of the third partition within at least one volatile memory component of the coordinator; and performing, at the coordinator, the first task with the third partition. The method may include, in response to a failure to receive the indication of completion of the first task with the second partition by the second node device within the predetermined time period: assigning the second task on the second partition to the first node device; and transmitting to the first node device an indication of the assignment of performing the second task on the second partition to cause the second partition to fetch the second partition.

The implementations of the first task with the first and second partitions may include using the first and second partitions as inputs to implement the first task to create corresponding partitions of a second data set; and performing the second task on the second partition may include using the second partition as an input to perform the second task to generate a corresponding partition of a third data set. The transmission of the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to: retrieve the first partition from one or more storage devices; Using at least a portion of the first partition stored within the volatile memory of the first node device as an input for performing the first task by the first node device; and transmitting the indication of completion of performing the first task with the first partition while leaving at least a portion of the first partition stored within the volatile memory of the first node device. The transmission of the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to: retrieve the second partition from one or more storage devices; Using at least a portion of the second partition stored within the volatile memory of the second node device as an input for performing the first task by the second node device; and transmitting the indication of completion of performing the first task with the second partition while leaving at least a portion of the second partition stored within the volatile memory of the second node device. The transmission of the indication of assignment of the second job on the second partition to the second node device may cause the second node device to use at least a portion of the second partition remaining within the volatile memory of the second node device as an input to perform the second second task is stored by the second node device to minimize accesses to the second partition stored within the non-volatile memory of the second node device.

The implementations of the first task with the first and second partitions may include implementations of the first task to generate the first and second partitions as outputs of the first task using corresponding partitions of a second data set as inputs; and performing the second task on the second partition comprises using the second partition as an input to perform the second task to generate a corresponding partition of a third data set. The transmission of the indication of assignment of performing the first task with the first partition to the first node device may cause the first node device to: generate the first partition as an output of performing the first task by the first node device; and transmitting the indication of completion of performing the first task with the first partition while leaving at least a portion of the first partition stored within the volatile memory of the first node device. The transmission of the indication of assignment of performing the first task with the second partition to the second node device may cause the second node device to: generate the second partition as an output of performing the first task by the second node device; and transmitting the indication of completion of performing the first task with the second partition while leaving at least a portion of the second partition stored within the volatile memory of the second node device. The transmission of the indication of assignment of the second job on the second partition to the second node device may cause the second node device to use at least a portion of the second partition remaining within the volatile memory of the second node device as an input to perform the second second task is stored by the second node device to minimize accesses to the second partition stored within the non-volatile memory of the second node device.

An apparatus may include a processor and memory for storing instructions that, when executed by the processor, cause the processor to perform operations including: analyzing a current status of resources of at least one node device of a plurality of node devices to determine availability of at least one graphics processing unit (GPU) of the at least one node device to be assigned to perform a first task of an analysis routine, wherein: the operation of the plurality of node devices is coordinated to at least partially perform tasks of analysis routines perform in parallel; the analysis routine is generated for execution by at least one central processing unit (CPU) of the at least one node; and the resources of the at least one node device are selected from a group consisting of the at least one CPU, the at least one GPU, and storage space within at least one memory of the at least one node device. In response to a determination that the at least one GPU is available to be assigned to perform the first task of the analysis routine, the processor may perform operations including: analyzing a first task routine of the analysis routine to determine if the first task routine is compiling may be to generate a GPU task routine for execution by the at least one GPU to cause the at least one GPU to at least partially parse multiple instances of the first task of the parsing routine without a dependency between inputs and outputs of the multiple instances of the first task wherein: the first task routine is generated for execution by the at least one CPU to perform the first task of the analysis routine; and determining whether the first task routine can be compiled to generate the GPU task routine, determining whether the first task routine includes an instruction that prohibits compilation to generate the GPU task routine, and a determination of whether inputs and outputs of the first task routine are defined so as not to require dependency. In response to a determination that the first task routine can be compiled to generate the GPU task routine, the processor may perform operations including: assigning a record partition of a plurality of record partitions of a record to the at least one node device to access the record partition to allow the at least one GPU; Using a conversion rule to translate at least one instruction of the first task routine into at least one corresponding instruction of the GPU task routine; Compiling the at least one corresponding instruction of the GPU task routine for execution by the at least one GPU; and assigning execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the plurality of instances of the first task with the record partition by the at least one GPU.

To determine whether the first task routine includes an instruction that prevents compilation to generate the GPU task routine, the processor may be caused to: determine whether the instruction of the first task routine is included in a set of instructions that are not can be converted into at least one instruction that can be executed by the at least one GPU; and in response to a determination that the instruction of the first task routine is not included in the set of instructions, determining whether the instruction of the first task routine in the first task routine is used in a manner that prevents conversion to at least one of the instructions the at least one GPU can be executed. To convert the at least one instruction of the first task routine into the at least one corresponding instruction of the GPU task routine, the processor may be caused to change the at least one instruction of the first task routine from a first programming language to the at least one corresponding instruction in a second programming language according to the conversion rule convert. The at least one memory of the at least one node device may include a first volatile memory communicatively coupled to the at least one CPU and a second volatile memory communicatively coupled to the at least one GPU; assigning the record partition to the at least one node device to allow access to the record partition by the at least one GPU may include causing the record partition to be stored within the second volatile memory; and in response to a determination that the at least one GPU is not available to be assigned to perform the first task of the analysis routine, the processor may be caused to perform operations including: omitting the analysis of the first task routine to determine whether the first task routine can be compiled to generate the GPU task routine; Assigning the record partition to the at least one node device to cause storage of the record partition within the first volatile memory to allow access to the record partition by the at least one CPU; Compiling the first task routine for execution by the at least one CPU; and assigning the performing the first task of the analysis routine with the record partition to the at least one node device to enable the first task with the record partition to be performed by the at least one CPU.

The apparatus may include a coordinator that coordinates the operation of the plurality of node devices; the processor may be caused to periodically receive updates to the current status of each node device of the plurality of node devices; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the processor may be caused to identify a node device of the plurality of node devices that integrates a GPU indicated by the current status as available becomes. To assign the record partition of the record to the at least one node device, the processor may be caused to perform operations including: analyzing metadata indicative of structural features of the record to identify a constraint in a manner that the record in the plurality of data set partitions may be partitioned, wherein the restriction is selected from a group consisting of a display of a minimum atomic unit of data within the data set and a specification of a partitioning scheme; and deriving a division of the record into the plurality of record partitions based at least in part on the constraint. The processor may be caused to perform operations, including: retrieving the metadata from at least one storage device in which the record is stored; and transmitting an indication of the assignment of the record partition to the at least one node device or the at least one storage device to cause a transfer of the record partition from the at least one storage device to the at least one node device.

The apparatus may include a node device of the at least one node device; the node device may comprise a GPU of the at least one GPU; the processor may include a CPU of the at least one CPU; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the CPU may be made to determine whether the GPU of the node device is indicated as being available by the current status. The processor is caused to perform operations including analyzing a second task routine of the analysis routine to determine if the second task routine can be compiled to generate another GPU task routine for execution by the at least one GPU to cause the second task routine to execute at least one GPU executes at least partially in parallel multiple instances of the second task of the analysis routine without a dependency between inputs and outputs of the multiple instances of the second task, the second task routine being generated for execution by the at least one CPU to perform a second task of the analysis routine. In response to a determination that the second task routine can not be compiled to generate the other GPU task routine, the processor may perform operations including: compiling the second task routine for execution by the at least one CPU; and assigning the second task of the analysis routine with the record partition to the at least one node device to enable the second task with the record partition to be performed by the at least one CPU.

The conversion rule may be selected from a group consisting of: a specification of a set of instructions, each of which prevents the compilation of the first task routine to generate the GPU task routine, if present within the first task routine; a specification of a set of instructions, each of which would not prevent the compilation of the first task routine to generate the GPU task routine, if present within the first task routine; specifying a way of using an instruction that prevents compilation of the first task routine to generate the GPU task routine if the manner of using the instruction occurs within the first task routine, otherwise the presence of the instruction within the first task routine does not Prevents compilation of the first task routine for generating the GPU task routine; a specification of a procedure for converting instructions in the first task routine to be looped by the at least one CPU into corresponding instructions of the GPU task routine executed in parallel by the at least one GPU in a corresponding loop with fewer iterations than the one Loop to be executed; a specification of a procedure for converting instructions in the first task routine to be looped by the at least one CPU into corresponding instructions of the GPU task routine to be executed in parallel by the at least one GPU and not in a loop; and a specification of a procedure for transforming instructions in the first task routine defining a data structure comprising entries to be sequentially accessed by the at least one CPU during execution of the first task routine, into corresponding instructions of the GPU task routine containing a define corresponding data structure comprising entries to be accessed in parallel during execution of the GPU task routine by the at least one GPU.

A computer program product concretely embodied in a non-transitory machine-readable storage medium, the computer program product comprising instructions operable to cause a processor to perform operations including: analyzing a current status of resources of at least one node device of a plurality of Node devices for determining availability of at least one graphics processing unit (GPU) of the at least one node device to be assigned to perform a first task of an analysis routine, wherein: the operation of the plurality of node devices is coordinated to perform tasks of the Perform analysis routine at least partially in parallel; the analysis routine is generated for execution by at least one central processing unit (CPU) of the at least one node; and the resources of the at least one node device are selected from a group consisting of the at least one CPU, the at least one GPU, and storage space within at least one memory of the at least one node device. In response to a determination that the at least one GPU is available to be assigned to perform the first task of the analysis routine, the processor may be caused to perform operations including: analyzing a first task routine of the analysis routine to determine if the The first task routine may be compiled to generate a GPU task routine for execution by the at least one GPU to cause the at least one GPU to at least partially parse multiple instances of the first task of the parsing routine without a dependency between inputs and outputs of the multiple instances performing the first task, wherein: the first task routine is generated for execution by the at least one CPU to perform the first task of the analysis routine; and determining whether the first task routine can be compiled to generate the GPU task routine, determining whether the first task routine includes an instruction that prohibits compilation to generate the GPU task routine, and a determination of whether inputs and outputs of the first task routine are defined so as not to require dependency. In response to a determination that the first task routine can be compiled to generate the GPU task routine, the processor may be caused to perform operations including: assigning a record partition of a plurality of record partitions of a record to the at least one node device for access to enable the record partition by the at least one GPU; Using a conversion rule to translate at least one instruction of the first task routine into at least one corresponding instruction of the GPU task routine; Compiling the at least one corresponding instruction of the GPU task routine for execution by the at least one GPU; and assigning execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the plurality of instances of the first task with the record partition by the at least one GPU.

To determine whether the first task routine includes an instruction that prevents compilation to generate the GPU task routine, the processor may be caused to: determine whether the instruction of the first task routine is included in a set of instructions that are not can be converted into at least one instruction that can be executed by the at least one GPU; and in response to a determination that the instruction of the first task routine is not included in the set of instructions, determining whether the instruction of the first task routine in the first task routine is used in a manner that prevents conversion to at least one of the instructions the at least one GPU can be executed. To convert the at least one instruction of the first task routine into the at least one corresponding instruction of the GPU task routine, the processor may be caused to change the at least one instruction of the first task routine from a first programming language to the at least one corresponding instruction in a second programming language according to the conversion rule convert. The at least one memory of the at least one node device comprises a first volatile memory communicatively coupled to the at least one CPU and a second volatile memory communicatively coupled to the at least one GPU; assigning the record partition to the at least one node device to allow access to the record partition by the at least one GPU may include causing the record partition to be stored within the second volatile memory; and in response to a determination that the at least one GPU is not available to be assigned to perform the first task of the analysis routine, the processor is caused to perform operations, including: omitting the analysis of the first task routine to determine whether the first task routine can be compiled to generate the GPU task routine; Assigning the record partition to the at least one node device to cause storage of the record partition within the first volatile memory to allow access to the record partition by the at least one CPU; Compiling the first task routine for execution by the at least one CPU; and assigning the performing the first task of the analysis routine with the record partition to the at least one node device to enable the first task with the record partition to be performed by the at least one CPU.

The processor may be a component of a coordinator that coordinates the operation of the plurality of node devices; the processor may be caused to periodically receive updates to the current status of each node device of the plurality of node devices; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the processor may be caused to identify a node device of the plurality of node devices that integrates a GPU that is displayed as available in the current state , In order to assign the record partition of the record to the at least one node device, the processor is caused to perform operations including: analyzing metadata indicative of structural features of the record to identify a constraint in a manner that the record falls within A plurality of data set partitions, wherein the restriction is selected from a group consisting of a display of a minimum atomic unit of data within the data set and a specification of a partitioning scheme; and deriving a division of the record into the plurality of record partitions based at least in part on the constraint. The processor may be caused to perform operations, including: retrieving the metadata from at least one storage device in which the record is stored; and transmitting an indication of the assignment of the record partition to the at least one node device or the at least one storage device to cause a transfer of the record partition from the at least one storage device to the at least one node device.

The processor may include a CPU of the at least one CPU; the CPU may be a component of a node device of the at least one node device; the node device may comprise a GPU of the at least one GPU; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the CPU may be made to determine whether the GPU of the node device is indicated as being available by the current status. The processor may be caused to perform operations including: analyzing a second task routine of the analysis routine to determine if the second task routine can be compiled to generate another GPU task routine for execution by the at least one GPU to cause that the at least one GPU performs at least partially in parallel multiple instances of the second task of the analysis routine without a dependency between inputs and outputs of the multiple instances of the second task, the second task routine being executed for execution by the at least one CPU to perform a second task of the analysis routine perform. In response to a determination that the second task routine can not be compiled to generate the other GPU task routine, the processor may be caused to: compile the second task routine for execution by the at least one CPU; and assigning the second task of the analysis routine with the record partition to the at least one node device to enable the second task with the record partition to be performed by the at least one CPU. The at least one GPU may support the execution of the at least one corresponding instruction of the GPU task routine in parallel over at least one thousand threads of execution.

A computer-implemented method may include: analyzing a current status of resources of at least one node device of a plurality of node devices to determine availability of at least one graphics processing unit (GPU) of the at least one node device to be assigned a first one Perform an analysis routine, wherein: the operation of the plurality of node devices is coordinated to perform at least partially parallel analysis task tasks; the analysis routine is generated for execution by at least one central processing unit (CPU) of the at least one node; and the resources of the at least one node device are selected from a group consisting of the at least one CPU, the at least one GPU, and storage space within at least one memory of the at least one node device. The method, in response to a determination that the at least one GPU is available to be assigned to perform the first task of the analysis routine, comprises: analyzing a first task routine of the analysis routine to determine if the first task routine can be compiled; to generate a GPU task routine for execution by the at least one GPU to cause the at least one GPU to perform at least partially in parallel multiple instances of the first task of the analysis routine without a dependency between inputs and outputs of the multiple instances of the first task the first task routine for execution is generated by the at least one CPU to perform the first task of the analysis routine; and determining whether the first task routine can be compiled to generate the GPU task routine, determining whether the first task routine includes an instruction that prohibits compilation to generate the GPU task routine, and a determination of whether inputs and outputs of the first task routine are defined so as not to require dependency. The process can be in response to a determination that the first task routine may be compiled to generate the GPU task routine comprises: assigning a record partition of a plurality of record partitions of a record to the at least one node device to allow access to the record partition by the at least one GPU; Using a conversion rule to translate at least one instruction of the first task routine into at least one corresponding instruction of the GPU task routine; Compiling the at least one corresponding instruction of the GPU task routine for execution by the at least one GPU; and assigning execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the plurality of instances of the first task with the record partition by the at least one GPU.

The determination of whether the first task routine includes an instruction that prevents compilation to produce the GPU task routine may include: determining whether the instruction of the first task routine is included in a set of instructions that are not converted into at least one instruction can be performed by the at least one GPU; and in response to a determination that the instruction of the first task routine is not included in the set of instructions, determining whether the instruction of the first task routine in the first task routine is used in a manner that prevents conversion to at least one of the instructions the at least one GPU can be executed. The converting the at least one instruction of the first task routine into the at least one corresponding instruction of the GPU task routine may comprise converting the at least one instruction of the first task routine from a first programming language into the at least one corresponding instruction in a second programming language according to the conversion rule.

The at least one memory of the at least one node device may include a first volatile memory communicatively coupled to the at least one CPU and a second volatile memory communicatively coupled to the at least one GPU; assigning the record partition to the at least one node device to allow access to the record partition by the at least one GPU may include causing the record partition to be stored within the second volatile memory; and in response to a determination that the at least one GPU is not available to be assigned to perform the first task of the analysis routine, the method may include: omitting the analysis of the first task routine to determine if the first task routine is being compiled can to generate the GPU task routine; Assigning the record partition to the at least one node device to cause storage of the record partition within the first volatile memory to allow access to the record partition by the at least one CPU; Compiling the first task routine for execution by the at least one CPU; and assigning the performing the first task of the analysis routine with the record partition to the at least one node device to enable the first task with the record partition to be performed by the at least one CPU.

A coordinator may coordinate the operation of the plurality of node devices; the coordinator may periodically receive updates to the current status of each node device of the plurality of node devices; and analyzing the current status to determine the availability of the at least one GPU of the at least one node device may include identifying at the coordinating device a node device of the plurality of node devices that integrates a GPU that is displayed as available in the current state. Assigning the record partition of the record to the at least one node device may comprise: analyzing, at the coordinator, metadata indicative of structural features of the record to identify a constraint in a manner in which the record into the plurality of record partitions wherein the constraint is selected from a group consisting of a display of a minimum atomic unit of data within the record and a specification of a partitioning scheme; and deriving a division of the record into the plurality of record partitions based at least in part on the constraint. The method may include: retrieving the metadata from at least one storage device in which the record is stored; and transmitting an indication of the assignment of the record partition to the at least one node device or the at least one storage device to cause a transfer of the record partition from the at least one storage device to the at least one node device.

Analyzing the current state to determine the availability of the at least one GPU of the at least one node device may include analyzing a current status of the node device by a CPU of a node device, whether a GPU of the node device is up-to-date wherein: the at least one CPU may comprise the CPU; the at least one node device may comprise the node device; and the at least one GPU may include the GPU.

The method of claim 21, comprising: analyzing a second task routine of the analysis routine to determine if the second task routine can be compiled to generate another GPU task routine for execution by the at least one GPU to cause the at least one GPU task routine to execute a GPU executes at least partially in parallel a plurality of instances of the second task of the analysis routine without a dependency between inputs and outputs of the multiple instances of the second task, the second task routine being generated for execution by the at least one CPU to perform a second task of the analysis routine. The method may include, in response to a determination that the second task routine can not be compiled to generate the other GPU task routine: compile the second task routine for execution by the at least one CPU; and assigning execution of the second task of the analysis routine with the record partition to the at least one node device to enable the second task with the data set partitions to be performed by the at least one CPU. The method may include analyzing the analysis routine to determine a sequence of tasks of the analysis routine, the order of the tasks comprising a relative order of the first and second tasks.

The foregoing will become more apparent together with other features and embodiments with reference to the following description, claims and appended drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure will be described in conjunction with the accompanying drawings:

1 FIG. 12 illustrates a block diagram that provides an illustration of the hardware components of a computing system, in accordance with some embodiments of the present technology.

2 FIG. 3 illustrates an example network that includes an example set of devices that communicate with each other through an exchange system and over a network, in accordance with some embodiments of the present technology.

3 FIG. 12 illustrates an illustration of a conceptual model of a communication protocol system according to some embodiments of the present technology. FIG.

4 FIG. 12 illustrates a communication grid computing system including a plurality of control and worker nodes, in accordance with some embodiments of the present technology.

5 FIG. 12 illustrates a flowchart depicting an example process for adjusting a communication grid or work project in a communication grid after a node has failed according to some embodiments of the present technology.

6 Figure 12 illustrates a portion of a communication grid computing system including a control node and a worker node, in accordance with some embodiments of the present technology.

7 FIG. 12 illustrates a flowchart illustrating an example process for performing a data analysis or processing project, in accordance with some embodiments of the present technology.

8th FIG. 12 illustrates a block diagram including components of an Event Stream Processing Engine (ESPE) according to embodiments of the present technology.

9 FIG. 12 illustrates a flowchart depicting an example process including operations performed by an event stream processing engine, in accordance with some embodiments of the present technology.

10 FIG. 12 illustrates an ESP system connected via an interface between a publishing device and multiple event subscribing devices in accordance with embodiments of the present technology. FIG.

11A and 11B each illustrate an embodiment of a distributed processing system.

12 illustrates an embodiment of a page exchange.

13A and 13B each illustrate an embodiment of assigning performance of tasks of an analysis routine to node devices of a distributed processing system.

14A . 14B and 14C together illustrate an embodiment of assigning record partitions and tasks.

15A and 15B together illustrate an example of performance of an assigned task by an embodiment of a node device.

16 illustrates another embodiment of assigning record partitions and tasks.

17A . 17B and 17C together illustrate an embodiment of delaying the allocation of a record partition and a task.

18A . 18B and 18C together illustrate another embodiment of delaying the allocation of a record partition and a task.

19 illustrates yet another embodiment of an analysis of task routines and assignment of tasks by a coordinator.

20 illustrates yet another embodiment of an analysis and compilation of a task routine by a node device.

21 illustrates yet another embodiment of initial and subsequent assignments of tasks by a coordinator.

22 Figure 12 illustrates yet another embodiment of deriving a time period of a delay for use in a subsequent assignment of a task.

23 illustrates yet another embodiment of a delayed assignment of a task by a coordinator.

DETAILED DESCRIPTION

Various embodiments described herein are generally directed to techniques to improve the effective use of processing, memory and network bandwidth resources within a grid of node devices to enable individual analysis routines to be completed in less time. Various embodiments are also generally directed to techniques for improving the ease of using multithreading support provided by at least some processors of a grid. Each node device of a grid of node devices may include both volatile memory that provides faster access to data and routines, and non-volatile memory that provides higher storage capacity but slower access than the volatile memory. In addition, each node of the grid may include one or more central processing units (CPUs), or may comprise a combination of one or more CPUs and one or more graphics processing units (GPUs), each of which may be more appropriate. to perform different types of tasks of an analysis routine. The initial assignments of data set partitions and task (s) of an analysis routine to node devices may be based on the availability of resources within the node devices and / or automated analysis of the task routines within the analysis routine to identify task routines that are used to execute GPUs instead of CPUs can be converted and compiled. Such an analysis may be performed either in connection with deriving the initial assignments or as part of compiling each of the task routines for execution by CPUs and / or by GPUs. Then, during the execution of task routines of the analysis routine, when a node device becomes available, it may become a next task for Assignment to a particular record partition, such assignment to the one node device is delayed by a predetermined period of time to give another node device, which still performs a previous task with that particular record partition, a chance to become available and assigned to perform this next task with this particular record partition. Such a delay in the assignment may allow the time-limited storage of the particular record partition within the volatile memory of the other node device to be used to thereby more quickly enable the execution of the next task with the particular record partition. Such a delay in assignment may also avoid the occurrence of a potentially larger delay associated with the transmission of the particular record partition to the one node device.

A grid device coordinator of node devices may periodically receive node data from each of the nodes of the node device grid, with periodically updated indications of the amount of availability of various processing, storage, and / or network access resources being provided in each one. The coordinator may also receive an analysis routine comprising executable instructions for a plurality of task tasks to be performed on at least one record and an order in which the tasks are to be performed. The coordinator may further receive metadata indicative of various structural features of at least the one record. From the node data, the metadata and / or the analysis routine, the coordinator may derive initial assignments of record partitions of the at least one record to selected ones of the node devices and initial assignments of the task (s) to be performed by the selected node devices. The coordinator may then transmit indications of the initial assignments to the selected node devices. As part of such initial assignments, multiple record partitions of at least the one record may be distributed among the selected node devices to allow the selected node devices to perform the same tasks at least partially in parallel with their respective record partitions. In addition, the coordinator may transmit to either a single storage device or a grid of storage devices that stores the data set displaying the distribution of data set partitions to be made to the selected node devices as part of the initial allocations to enable the distribution device Start executions of the initial task (s). In some embodiments, each of the selected node devices may retrieve one or more of the record partitions from the one or more of the storage devices. In other embodiments, one or more of the record partitions may be transmitted to each of the selected node devices through the one or more of the storage devices.

With respect to the processing resources over which the coordinator can periodically receive node data, each of the node devices may integrate one or more GPUs in addition to or instead of integrating one or more CPUs. The one or more CPUs may use an internal processing architecture that is found to be well suited for serial processing of task routines involving various input / output operations and / or branch operations that require the execution of different sets of instructions on the results of different determinations , The one or more CPUs may each include one or more processing cores, each of which may support a relatively limited degree of parallel execution of instructions on a relatively limited number of execution threads. In contrast, the one or more GPUs may use an internal processing architecture that is found to be well-suited for the parallel processing of task routines that involve a relatively limited variety of computations and / or bitwise operations. In some embodiments, the one or more GPUs may be capable of supporting the parallel processing of a relatively large set of instances of a task over a relatively large number of threads of execution if there are no dependencies between the instances of the task (sometimes referred to as "Embarrassingly parallel"). In fact, for a relatively limited variety of tasks, a single GPU within a single node device may be able to perform the same work as the CPUs of numerous separate node devices, but faster and more cost effectively. Thus, it may be advantageous to have at least a subset of the node devices that integrate one or more GPUs capable of performing such a limited variety of tasks with such an increase in speed, and it may be advantageous to do so be to automatically identify tasks in analysis routines that are of such a limited variety.

In embodiments of a node device grid in which some, but not all, of the node devices integrate such GPUs, whether an analysis routine includes task routines for one or more tasks accessible to be executed by GPUs as scrupulously parallel tasks faster than CPUs, the coordinator device will determine whether priority is given to the allocation of node devices that integrate GPUs or node devices that do not integrate GPUs to perform the tasks of the analysis routine. The coordinator may analyze the task routines of an analysis routine to identify tasks that are implemented with instructions and that operate on data in a manner that avoids dependencies that make those tasks accessible thereto, for execution as embarrassing parallel tasks over one large number of threads to be compiled by one or more GPUs. If no such task routine is found by such an analysis, then the set of analysis routines can be compiled for execution only by CPUs. Also, priority may be given to the assignment of tasks of the analysis routines to be performed by node devices that do not integrate GPUs and / or integrate the one or more CPUs and one or more GPUs. Such prioritization may be effected to leave as many of the nodes integrating one or more GPUs as available as possible to be assigned to tasks of another analysis routine, in which the instructions and / or interaction with data in one or more Task routines are available for execution as embarrassing parallel tasks to be compiled by GPUs.

It should be noted, however, that although the results of the analysis of the task routines of an analysis routine may have some influence on which node devices are selected for the assignment of tasks, in some embodiments, indications of which processing resources are available among the available node devices is one may have some influence on whether the task routines are parsed and / or compiled for execution by GPUs. In particular, in embodiments where none of the node devices that integrate GPUs are currently available to be assigned to any task (eg, all node devices that integrate GPUs are assigned to perform tasks of a different and completely unrelated analysis routine). For example, the coordinator may refrain from performing any analysis of the task routines to determine if any of the task routines are accessible to compile for execution by a GPU, as currently no GPUs are available for this purpose.

When performing an analysis of the instructions within task routines, such an analysis may at least not prevent comparisons of instructions for each task routine with a list of instructions that are each known to perform their respective tasks as scrupulously parallel tasks by a GPU , and / or with a list of instructions, each known to reflect a task incapable of being performed as a scrupulously parallel task by a GPU. In addition, if a task routine does not include instructions that render its corresponding task as unable to be performed as a scrupulously parallel task by a GPU, the manner in which the instructions within that task routine are used may be analyzed to determine whether the manner in which any instructions are used will render the corresponding task incapable of being performed as a scrupulously parallel task by a GPU. By way of example, if such instructions are used to perform operations on data in a manner that would create dependencies between instances of a task routine such that these instances could not actually be performed in parallel, then the task routine task may not be accessible to it an embarrassingly parallel task to be performed. If it is determined that the instructions of a task routine and the manner in which these instructions are used do not prevent the corresponding task from being performed as a scrupulously parallel task by a GPU, then compiling the instructions of the task routine may be for such execution be the standard procedure through one or more GPUs. If the node device grid includes node devices that integrate different GPUs that do not share an instruction set, compilation of the task routine for such execution by one or more GPUs may result in multiple compilations of the task routine to support each of the different GPUs.

In some embodiments, compiling a task routine to perform a task that is determined to be operable as a scrupulously parallel task by a GPU may involve a conversion of instructions of the task routine that were not generated Such embarrassing parallel execution of the task by a GPU to result in instructions that are generated to cause such a performance of the task. A compiler to do such a transformation can use a set of compile rules, each associated with one or more particular instructions that may be present under the instructions of the task routine and that cause the one or more particular instructions to be converted to one or more other instructions that cause scrupulous parallel execution by a GPU , Among such compile rules may be rules that each cause the conversion of a particular type of loop from instructions to another type of loop that is more amenable to scrupulously parallel execution by a GPU (eg, a loop with a reduced number of iterations ), and / or which cause replacement of a particular type of loop of instructions with one or more other instructions for execution by a GPU in an embarrassingly parallel manner that does not result in the use of a loop. There may also be a set of compile rules, each associated with a particular type of data structure that can be instantiated or otherwise used by the instructions of the task routine, and that cause the data structure to be converted to another data structure used for the use in an embarrassingly parallel execution by a GPU is more accessible.

By using such conversions of instructions of one or more task routines, personnel writing the instructions of the task routines of an analysis routine may be given the opportunity to utilize the embarrassingly parallel processing capabilities of the one or more GPUs integrated in at least some of the node devices. without the need to write the instructions of the task routines specifically for embarrassing parallel execution by GPUs. In other words, this staff may be spared the need to acquire the skills to architect and write the instructions that implement the tasks of an analysis routine in a manner designed for scrupulously parallel execution by GPUs. In addition, if the node gadget includes node devices that integrate different GPUs that do not share a set of instructions, then that staff will also be spared the need to architect and write different versions of the instructions of the task routines to handle the different characteristics of the embarrassingly parallel execution of each of the different ones To address GPUs.

With respect to the memory resources over which the coordinator can periodically receive node data, each of the node devices can integrate memory capabilities implemented as a combination of volatile and non-volatile memory. The volatile memory may be implemented with one or more memory components employing memory technology that provides relatively quick access to data and / or routines, but is unable to store data and / or routines stored therein without a continuous supply of electrical power to keep. Such technologies include, but are not limited to, any of a variety of types of Random Access Memory (RAM). The non-volatile memory may be implemented with one or more memory components employing memory technology capable of retaining data and / or routines stored therein, whether or not electrical power is still provided It is able to provide access as fast as that provided by various volatile storage technologies on which the volatile memory can be based. Such non-volatile memory technologies include the use of any of a variety of ferromagnetic and / or optical storage media.

Because of the faster access provided by the volatile memory as compared to the non-volatile memory, instructions that are being executed by the one or more CPUs and / or the one or more GPUs integrated into each node device may be executed , stored within volatile memory, where they can be read, written and / or modified faster. However, because of the often lower cost and / or higher storage densities of the non-volatile memory components compared to the volatile memory components, the non-volatile memory may be implemented with a higher storage capacity than the volatile memory within each of the node devices. Although there may be records that are sufficiently small and / or that are distributed among a sufficient number of node devices to cause each record partition of the record that is distributed to a node device to be sufficiently small to be completely within volatile Memory to be stored, it is envisaged that the data set partitions of the majority of data sets are each rather too large to do so.

As a result, within each node device, pages of routines being executed and / or data being accessed by the one or more CPUs and / or the one or more GPUs may be swapped from non-volatile memory to volatile memory , As may be known to those skilled in the art, any of a variety of algorithms can be used to select pages of routines and / or data that are swapped into volatile memory and / or to select pages to be held in volatile memory while others are swapped back to non-volatile memory, including, but not limited to, any of a variety of demand-based and / or predictive algorithms. In one or more embodiments, one or more of the node devices may execute an operating system (OS) that includes a paging component that performs such exchanging of uniformly sized pages of routines and / or data. Depending on various factors, such as the types of operations performed, the frequency of accesses to different pages of routines and / or data, and / or the number of routines executed in parallel, one page may be a routine and / or data within a volatile memory for a longer or shorter period of time before being swapped back to the non-volatile memory to free space within the volatile memory for another side of a routine and / or data. Thus, the storage of pages of routines and / or data within a volatile memory within each of the node devices may be time limited.

In node devices that integrate both one or more CPUs and one or more GPUs, there may be volatile memory for the one or more CPUs and another volatile memory for the one or more GPUs. However, there may be a single non-volatile memory, and pages of routines and / or data may be exchanged between the single non-volatile memory and each of the two volatile memories. In some embodiments, the operation of the one or more GPUs may be at least partially controlled by the one or more CPUs such that the one or more GPUs may not be fully autonomous from the one or more CPUs. In such embodiments, the volatile memory associated with the one or more GPUs may also be accessible to the one or more CPUs, and a memory page management routine executed by the one or more CPUs to swap pages from Routines and / or data for the one or more CPUs may also perform such swapping of pages of routines and / or data for the one or more GPUs. As those skilled in the art may appreciate, such replacement by the one or more CPUs on behalf of the one or more GPUs may occur due to the need for one or more driver routines to be executed by the one or more CPUs to access to allow the non-volatile memory and / or to use a file system used in storing data and / or routines as files with the non-volatile memory. Thus, regardless of the exact manner in which each of the selected node devices is provided with a record partition, such a received record partition may initially be completely stored within the non-volatile memory within each node device. After such receipt and storage, pages of the received record partition may then be swapped as needed into the volatile memory of the one or more CPUs and / or the one or more GPUs to assist in performing one or more tasks of an analysis routine with the record partition.

With respect to the network access resources over which the coordinator can periodically receive node data, each of the node devices may integrate a network interface to a network used by the node device grid to communicate with the node devices communicatively with each other, with the coordinator, and / or with one or more memory devices ( eg a memory device grid). The task routines executed by the CPU (s) and / or the GPU (s) to perform analysis routine tasks may be distributed by the coordinator to node devices over the network. Also, the record partitions with which the tasks are performed may be transferred from the one or more storage devices to the node devices over the network, and record partitions derived within the node may be transmitted back to the one or more storage devices. As may be known to those skilled in the art, in some embodiments, the one or more GPUs of a node device may not be able to directly operate the network interface of the node device to effect an exchange of routines and / or data to be performed a task by which one or more GPUs are associated. Instead, in a similar manner as exchanging pages associated with the one or more GPUs between volatile and non-volatile memory, the one or more CPUs of the node device may operate the network interface on behalf of the one or more GPUs. As previously mentioned, such an action on behalf of the one or more GPUs may be due to the need of one or more CPU routines to be executed by the one or more CPUs to allow access to the network interface.

Just as accessing routines and / or data stored within a non-volatile memory of a node device may be considerably slower than accessing routines and / or data stored within a volatile memory, access to routines may be and / or data stored within another device may be much slower across a network. Additionally, in some embodiments, accessing routines and / or data stored either within nonvolatile memory or within another device may even be slower across the network for the one or more GPUs due to their dependency on the one or more CPUs the node device to take action to allow such access on behalf of the one or more GPUs. Thus, it may be considered desirable to always maximize accesses made to routines and / or data while still stored within a volatile memory associated with the CPU (s) and / or GPU (s) making these accesses and minimizing accesses to routines and / or data while stored in non-volatile memory and / or within other devices such that access must occur over the network. This may result in some node devices of the node device grid becoming idle for different time periods to await the availability of particular node devices for use in performing certain tasks with particular record partitions rather than immediately assigning tasks to each node available for use in performing a task regardless of which node devices already have specific record partitions within their memory. In other words, the assignment of a next task may be delayed for a period of time to allow a particular node device in which a particular record partition is stored to become available again to be assigned a next task involving the use of the particular record partition instead of immediately allocating the next task to another node device to which the particular record partition would need to be transferred. In addition, if there are multiple tasks to be performed on a particular partition that can be performed using a GPU, delaying the assignment of the next of these multiple tasks may allow for the ability to assign this next of these tasks to a node device , which integrates a GPU and already stores the respective record partition, avoid a situation where the immediate assignment of the next task to the next available node device may result in an allocation to a node device that does not integrate a GPU, thus resulting in slower performance the next of these tasks.

Thus, for example, if a first node device that was busy performing a first task of analysis with a first record partition of a record completes its execution of the first task with the first record partition and thereby becomes available so that it could be assigned a second task of the analysis with a second record partition of the record, the coordinator may refrain from assigning the first node device to perform the second task with the second record partition for a predetermined period of time. The predetermined time period may be selected to provide an opportunity for a second node device that is still busy performing the first task with the second record partition to complete its performance of the first task with the second record partition, such that the second node device in FIG is able to perform the second task with the second record partition. However, if the second node device becomes unavailable to be assigned to perform the second task with the second record partition, then that other execution device may be assigned to another node device in which the second record partition is not already stored (eg, the first one node device). Alternatively, in other embodiments, if there is another node device in which the second record partition has also been stored in preparation for using that other node device as a backup in response to a failure of a node device performing a task with the second record partition, and if the second node device becomes unavailable to be assigned to perform the second task with the second record partition, and if such other node device is currently available to be assigned, the other such node device is so assigned.

In this way, the fact can be utilized that the second data set partition is already stored within the volatile and / or non-volatile memory of the second node device, so that the second data set partition does not have to be exchanged between devices in order to carry out the second task with the enable second record partition by another node device. In other words, if the second task with the second record partition had been assigned to the first node device, then the second record partition would have had to be transferred from the second node device or from the one or more storage devices to the first node device. In addition to consuming the available bandwidth of the network and the Network interface of at least the first node device, the execution by the first node device of the second task would necessarily be delayed with the second data set partition until at least enough of the second data set partition would be received by the first node device to allow execution to commence.

As another example, each of a first node device that was busy performing a first task of an analysis routine to create a first record partition of a record and a second node device that was busy performing the first task may have a second record partition of the record, both complete their accomplishments of the first task. However, while the first node device may be available to be assigned to a different task, the second node device (as a result of the release of node devices among a plurality of unrelated analyzes) may be engaged in performing a task of an unrelated analysis routine, such that the second Node device may not yet be available to be assigned to perform a second task of the analysis routine with the second record partition that has generated the second node device itself. As previously noted, the coordinator may refrain from allocating the first node device to perform the second task with the second data set partition for a predetermined period of time. The predetermined time period may be selected to provide a way for the second node device to complete its execution of the unrelated analysis routine task so that the second node device becomes available to be assigned to perform the second task with the second data set partition. As already mentioned, the fact can be used in this way that the second data record partition is already stored within the volatile and / or non-volatile memory of the second node device, so that the second data record partition does not have to be exchanged between devices in order to carry out the second task with the second record partition by another node device.

With general reference to notations and nomenclature used herein, portions of the following detailed description may be presented in terms of program procedures executed by a processor of one or more networked machines. These procedural descriptions and representations are used by professionals in the field to most effectively convey the substance of their work to other professionals. A procedure is here and generally conceived as a self-consistent sequence of operations leading to a desired result. These operations are those that require physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical, magnetic or optical communications capable of being stored, transferred, combined, compared and otherwise manipulated. It proves convenient at times, principally for reasons of common usage, to refer to what is being communicated as bits, values, elements, symbols, characters, terms, numbers, or the like. It should be noted, however, that all of these and similar terms are associated with the corresponding physical quantities and are merely convenient labels applied to those quantities.

Furthermore, these manipulations are often referred to by terms such as adding or comparing, usually associated with mental operations performed by a human operator. However, no such ability of a human operator in any of the operations described herein is necessary or desirable in most cases forming part of one or more embodiments. Instead, these operations are machine operations. Useful machines for performing operations of various embodiments include machines that are selectively activated or configured by a routine stored in the one written in accordance with the teachings described herein, and / or include apparatus specifically for the required purpose are constructed. Various embodiments also relate to devices or systems for performing these operations. These devices may be specially designed for the required purpose or may include a general purpose computer. The required structure for a variety of these machines will be apparent from the description given.

Referring now to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding thereof. However, it may be obvious that the novel embodiments may be practiced without these specific details. In other instances, well-known structures and devices are disclosed in US Pat Block diagram shown to facilitate a description thereof. The intent is to cover all modifications, equivalents, and alternatives within the scope of the claims.

The systems illustrated in some of the figures may be provided in various configurations. In some embodiments, the systems may be configured as a distributed system in which one or more components of the system are distributed over one or more networks in a cloud computing system and / or a fog computing system.

1 FIG. 4 is a block diagram illustrating an illustration of the hardware components of a communications network. FIG 100 according to embodiments of the present technology. The data transmission network 100 is a specialized computer system that can be used to process large amounts of data that requires a large number of computer processing cycles.

The data transmission network 100 can also be a computing environment 114 include. The computing environment 114 may be a specialized computer or other machine within the communications network 100 received data processed. The data transmission network 100 also includes one or more network devices 102 , The network devices 102 may include client devices trying to work with the computing environment 114 to communicate. For example, the network devices 102 including data to the computing environment to be processed 114 send signals to the computing environment 114 send to control various aspects of the computing environment or the data it processes. The network devices 102 can with the computing environment 114 interact through a number of ways, such as one or more networks 108 , As in 1 can be shown the computing environment 114 one or more other systems. For example, the computing environment 114 a database system 118 and / or a communication grid 120 include.

In other embodiments, network devices may be streaming a large amount of data, either all at once or over a period of time (eg, using event stream processing (ESP), further described with reference to FIG 8th - 10 is described), to the computing environment 114 over the nets 108 provide. For example, network devices 102 Network computers, sensors, databases, or other devices include the data to the computing environment 114 transmit or otherwise provide. For example, network devices may include local area network devices, such as routers, hubs, switches, or other computer network devices. These devices may provide a variety of stored or generated data, such as network data or data specific to the network devices themselves. Network devices may also include sensors that monitor their environment or other devices to collect data relating to that environment or devices, and such network devices may provide data that they collect over time. Network devices may also include devices within the Internet of Things, such as devices within a home automation network. Some of these devices may be referred to as edge devices and may include edge calculation circuits. Data can be sent from network devices directly to the computing environment 114 or to data storage connected to the network, such as the data storage connected to the network 110 , to be transferred to storage, so that the data later from the computing environment 114 or other parts of the data transmission network 100 can be retrieved.

The data transmission network 100 can also have one or more data storage connected to the network 110 include. The data storage connected to the network 110 are used to transfer data from the computing environment 114 and any intermediate or final data generated by the computing system in non-volatile memory. In certain embodiments, the configuration allows the computing environment 114 however, their operations are performed so that intermediate and final data results can be stored only in volatile memory (eg, RAM) without storing intermediate or final data results in non-volatile types of memory (eg, hard disk) Need to become. This can be useful in certain situations, such as when the computing environment 114 Receives ad hoc queries from a user and when responses generated by processing large amounts of data must be generated on-the-fly. In this non-limiting situation, the computing environment can 114 be designed so that it retains the processed information within the memory, so that answers can be generated for the user at various levels of detail and a user is allowed to interrogate this information interactively.

Network-attached data stores can store a variety of different types of data, organized in a variety of different ways, and from a variety of different sources. For example, a network attached storage may include a different storage than the primary storage that resides within the computing environment 114 located directly by processors located therein. The data storage connected to the network may include, but is not limited to, secondary, tertiary or auxiliary storage such as large hard disks, servers, virtual storage. Storage devices may include portable or non-portable storage devices, optical storage devices, and various other media capable of storing and containing data. A machine-readable storage medium or computer-readable storage medium may include a non-transitory medium in which data may be stored and which does not include carrier waves and / or transitory electronic signals. Examples of a non-transitory medium may include, for example, a magnetic disk or a magnetic tape, optical storage media such as a compact disk or digital versatile disk, flash memory, memory or memory devices. A computer program product may include code and / or machine-executable instructions that may represent a procedure, function, subroutine, program, routine, subroutine, module, software package, class, or any combination of instructions, data structures, or program instructions , A code segment may be coupled by passing and / or receiving information, data, arguments, parameters or memory contents with another code segment or hardware circuit. Information, arguments, parameters, data, etc. may be transferred, forwarded, or transmitted by any suitable means, including, but not limited to, memory sharing, message passing, token passing, network transmission. Further, the data stores may hold a variety of different types of data. For example, data storage connected to the network 110 unstructured (eg, raw) data such as manufacturing data (e.g., a database containing records identifying products created with parameter data for each product, such as colors and models) or product sales databases (e.g. A database that contains individual records that identify details of individual product sales).

The unstructured data can be used by the computing environment 114 in various forms, such as in a flat file or a conglomeration of records, and may have data values and accompanying timestamps. The computing environment 114 can be used to analyze the unstructured data in a variety of ways to determine the best way to structure that data (eg, hierarchically) so that the structured data is tailored for a type of further analysis, that a user wants to perform on the data. For example, after processing, the unstructured time-stamped data may be aggregated temporally (eg, into daily time-period units) to generate time-series data and / or hierarchically structured according to one or more dimensions (eg, parameters, attributes, and / or variables) become. For example, data may be stored in a hierarchical data structure, such as in a ROLAP or MOLAP database, or may be stored in a different tabular form, such as in a flat hierarchical form.

The data transmission network 100 can also have one or more server farms 106 include. The computing environment 114 can send selected communications or data to the one or more server farms 106 or forward one or more servers within the server farms. The server farms 106 may be configured to provide information in a predetermined manner. For example, the server farms 106 access data to send in response to a communication. The server farms 106 may be separate from another device within the communications network 100 such as the computing environment 114 , be housed and / or may be part of a device or system.

The server farms 106 You can use a variety of different types of data processing as part of the data transmission network 100 host. The server farms 106 can use a variety of different data from network devices, from the computing environment 114 , from the cloud network 116 or received from other sources. The data may have been received from one or more sensors as inputs from a control database, or collected or received as inputs from an external system or device. The server farms 106 may assist in processing the data by transforming raw data into processed data based on one or more rules implemented by the server farms. For example, sensor data may be analyzed to determine changes in an environment over time or in real time.

The data transmission network 100 can also have one or more cloud networks 116 include. The cloud network 116 may include a cloud infrastructure system that provides cloud services. In certain embodiments, services may be provided by the cloud network 116 provided, a variety of services that are provided on request to the users of the cloud infrastructure system. The cloud network 116 is in 1 as with the computing environment 114 connected (and therefore with the computing environment 114 as a client or user), but the cloud network 116 can with any of the devices in 1 be connected or used by them. The services provided by the cloud network can be dynamically scaled to meet the needs of their users. The cloud network 116 may include one or more computers, servers, and / or systems. In some embodiments, the computers, servers, and / or systems that differentiate the cloud network differ 116 from the user's own computers, servers and / or systems on-site. For example, the cloud network 116 host an application, and a user can order and use the application on demand over a communications network, such as the Internet.

Although every device, every server and every system in 1 As shown as a single device, it should be understood that multiple devices may be used instead. For example, a set of network devices may be used to transmit various communications from a single user or the remote server 140 can include a server stack. As another example, data may be part of the computing environment 114 are processed.

Every communication within the data transmission network 100 (eg between client devices, between the servers 106 and the computing environment 114 or between a server and a device) may be via one or more networks 108 respectively. The nets 108 They may include one or more of a variety of different types of networks, including a wireless network, a wired network, or a combination of a wired and wireless network. Examples of suitable networks include the Internet, a personal area network, a local area network (LAN), a wide area network (WAN) or a wireless local area network (WLAN). A wireless network may include a wireless interface or a combination of wireless interfaces. As an example, a network may be in the one or more networks 108 a short distance communication channel, such as a Bluetooth or a Bluetooth low energy channel. A wired network may include a wired interface. The wired and / or wireless networks may be implemented using routers, access points, bridges, gateways or the like to network devices 114 to connect, as further referring to 2 is described. The one or more networks 108 may be fully integrated within, or may include, an intranet, an extranet, or a combination thereof. In one embodiment, communications between two or more systems and / or devices may be achieved through a secure communication protocol, such as Secure Sockets Warehousing (SSL) or Transport Warehouse Security (TLS). In addition, data and / or transaction details can be encrypted.

Some aspects may use the Internet of Things (IoT), where things (eg, machines, devices, phones, sensors) can be connected to networks and the data of those things within things and / or outside of things can be collected and processed. For example, the IoT may include sensors in many different devices, and high quality analysis may be used to identify hidden relationships and increase efficiencies. This can apply to both big data analytics and real-time analytics (such as ESP). This will be further explained below with reference to 2 described.

As mentioned earlier, the computing environment 114 a communication grid 120 and a transmission network database system 118 include. The communication grid 120 can be a grid-based computing system for processing large amounts of data. The transmission network database system 118 It can be used to manage, store, and retrieve large amounts of data sent to one or more data stores attached to the network 110 or other data stores located at different locations within the transmission network database system 118 be located, distributed and stored in it. The compute nodes in the grid-based computing system 120 and in the transmission network database system 118 can use the same processor hardware, such as processors, that are within the computing environment 114 reside, share.

2 FIG. 3 illustrates an example network that includes an example set of devices communicating with each other via an exchange system and over a network according to embodiments of the present technology. As mentioned earlier, any communication within the Data transmission network 100 via one or more networks. The system 200 includes a network device 204 , which is designed to work with a variety of types of client devices, such as the client devices 230 to communicate over a variety of types of communication channels.

As in 2 shown, the network device 204 communication over a network (eg, a cellular network via a base station 210 ) transfer. Communication can be via the base station 210 to another network device, such as network devices 205 - 209 to be guided. Communication can also be via the base station 210 to the computing environment 214 be directed. For example, the network device 204 Data either from its environment or from other network devices (such as network devices 205 - 209 ) and collect this data to the computing environment 214 transfer.

Although the network devices 204 - 209 in 2 each shown as a mobile phone, laptop computer, tablet computer, temperature sensor, motion sensor, and audio sensor, the network devices may be or include sensors that are sensitive to detecting aspects of their environment. For example, the network devices may include sensors such as, but not limited to, water sensors, power sensors, electrical current sensors, chemical sensors, optical sensors, pressure sensors, geographic or position sensors (eg, GPS), speed sensors, acceleration sensors, flow rate sensors. Examples of properties that may be detected include force, torque, load, strain, position, temperature, air pressure, fluid flow, chemical properties, resistance, electromagnetic fields, radiation, irradiation, proximity, acoustics, humidity, distance, speed, Vibrations, acceleration, electrical potential, electric current. The sensors may be attached to various components used as part of a variety of different types of systems (eg, an oil drilling operation). The network devices can recognize and record data related to the environment they are monitoring and that data to the computing environment 214 transfer.

As previously mentioned, one type of system, which may include various sensors collecting data to be processed and / or transmitted to a computing environment in accordance with certain embodiments, includes an oil drilling system. For example, the one or more boring operation sensors may measure surface sensors that measure a hook load, a fluid rate, a temperature and a density in and out of the borehole, a riser pressure, a surface torque, a drill pipe speed, a propulsion rate, a mechanical specific energy, etc. and borehole sensors that include a speed of a drill bit, fluid densities, borehole torque, borehole vibration (axial, tangential, lateral), a weight applied to a drill bit, a ring pressure, a differential pressure, an azimuth, a slope, a borehole kink, a measured depth, a vertical Depth, a well temperature, etc., include. In addition to the raw data collected directly from the sensors, other data may include parameters that are either developed by the sensors or assigned to the system by a client or other controller. For example, one or more drilling operation control parameters may control settings such as a mud engine speed to flow ratio, a drill tip diameter, a predicted formation peak, seismic data, weather data, and so on. Other data may be generated using physical models such as an earth model, a weather model, a seismic model, a bottom hole mounting model, a wellbore plan model, a ring friction model, and so on. In addition to sensor and control settings, predicted outputs may include, for example, rate of advance, mechanical specific energy, hook load, fluid rate flow, effluent fluid rate, pump pressure, surface torque, drill pipe speed, ring pressure, ring friction pressure, ring temperature , the equivalent circulation density, etc. are also stored in the data warehouse.

In another example, another type of system, which may include various sensors collecting data to be processed and / or transmitted to a computing environment according to certain embodiments, includes home automation or a similar automated network in another environment, such as in an office room, a school, a public space, a sports facility or a variety of other places. Network devices in such an automated network may include network devices that allow a user to access various home appliances that may be within the user's home (eg, a television, a radio, a light, a fan, a humidifier, a sensor, etc.) Microwave oven, iron and / or the like) or outside the user's home (eg, exterior motion sensors, exterior lights, garage door openers, sprinklers, or the like), to control, and / or configure them. For example, the network device 102 a home automation switch, which may be coupled to a household appliance. In another embodiment, a network device may enable a user Access, control, and / or configure devices such as office-related devices (eg, copier, printer, or fax), audio and / or video-related devices (e.g., a receiver, a speaker, a projector, a DVD player or a television), media players (eg, a compact disc player, a CD player, or the like), computing devices (eg, a home computer, a laptop computer, a tablet, a personal digital Assistant (PDA), a computing device or a portable device), lighting devices (eg, a lamp or built-in lights), devices associated with a security system, devices associated with an alarm system, devices that operate in a motor vehicle (eg, wireless devices, navigation devices), and / or the like. The data may be collected from such various sensors in raw form, or data may be processed by the sensors to produce parameters or other data either developed by the sensors based on the raw data or assigned to the system by a client or other controller are.

In another example, a type of system that may include various sensors that collect data that is to be processed and / or transmitted to a computing environment in accordance with certain embodiments includes a power grid. A variety of different network devices may be included in a power grid, including, but not limited to, various devices within one or more power plants, energy farms (eg, wind farm, solar park, among others), energy storage facilities, factories, homes, and consumer businesses. One or more of these devices may include one or more sensors that detect energy gain or loss, electrical input or output or loss, and a variety of other performance levels. These sensors can collect data to inform users about how the power grid and individual devices within the network work and how they can be made more efficient.

Network device sensors can also perform processing on data that is collected before the data is sent to the computing environment 114 or before deciding whether to send data to the computing environment 114 to be transferred. For example, network devices may determine whether the collected data meets certain rules, such as by comparing data or values calculated from the data, and comparing that data to one or more thresholds. The network device may use these data and / or comparisons to determine if the data is for further use or processing to the computing environment 214 to be transferred.

The computing environment 214 can the machines 220 and 240 include. Although the computing environment 214 in 2 with two machines, 220 and 240 , can be shown, the computing environment 214 have only one machine or more than two machines. The machines that make up the computing environment 214 may include specialized computers, servers, or other machines designed to handle large amounts of data individually and / or collectively. The computing environment 214 may also include storage devices that include one or more databases of structured data, such as data organized in one or more hierarchies, or unstructured data. The databases can be used with the processing devices within the computing environment 214 communicate to distribute data to them. Because network devices send data to the computing environment 214 can transfer this data from the computing environment 214 received and then stored within these storage devices. Data coming from the computing environment 214 can also be used in data stores 235 which are also part of the computing environment 214 be or be associated with it.

The computing environment 214 can work with different devices via one or more routers 225 or other inter-network or intra-network connection components. For example, the computing environment 214 with the devices 230 via one or more routers 225 communicate. The computing environment 214 may store data from or relating to communications, client device operations, client rules, and / or actions associated with the user stored in one or more data stores 235 are stored, collect, analyze and / or store. Such data may be the communication routing to the devices within the computing environment 214 like data within the computing environment 214 saved or processed and other actions.

In particular, various other devices may also be used to facilitate communication routing and / or processing between devices within the computing environment 214 and with devices outside the computing environment 214 to influence. For example, the computing environment 214 , as in 2 shown a web server 240 include. Thus, the computing environment 214 Retrieve data of interest, such as client information (e.g., product information, client rules, etc.), technical product details, news, current or predicted weather data, and so on.

In addition to that the computing environment 214 Collecting data (such as that received from network devices, such as sensors, and client devices or other sources) that are to be processed as part of a Big Data Analytics project can also be used as part of a streaming analytics environment Receive real time. As already mentioned, data can be collected using a variety of sources as communicated across different types of networks or locally. Such data may be received on a real-time streaming basis. For example, network devices may periodically receive data from network device sensors as the sensors continuously detect, monitor, and track changes in their environments. Devices within the computing environment 214 can also pre-analyze data received to determine whether the received data should be processed as part of an ongoing project. The data coming from the computing environment 214 regardless of what the source or method or timing of the reception is, they can be processed over a period of time to allow a client to determine result data based on the needs and rules of the client.

3 FIG. 12 illustrates an illustration of a conceptual model of a communication protocol system according to embodiments of the present technology. FIG. In particular identified 3 the operation of a computing environment in an Open Systems Interaction model that corresponds to different connection components. The model 300 shows, for example, how a computing environment, such as the computing environment 314 (or the computing environment 214 in 2 ) can communicate with other devices in their network and control how and under what conditions the communication between the computing environment and other devices is performed.

The model can be the layers 302 - 314 include. The layers are arranged in a stack. Each layer in the stack serves the layer that is one level higher than it (except the application layer, which is the highest layer), and is served by the layer that is one level lower (with the exception of the physical layer, which is the lowest) Layer is). The physical layer is the lowest layer because it receives and transmits raw bytes of data and is the farthest layer from the user in a communication system. On the other hand, the application layer is the highest layer because it interacts directly with a software application.

As already mentioned, the model includes a physical layer 302 , The physical layer 302 represents the physical communication and can define parameters of this physical communication. For example, such physical communication may take the form of electrical, optical or electromagnetic signals. The physical layer 302 also defines protocols that can control communication within a communications network.

The connection layer 304 defines connections and mechanisms used to transmit (ie move) data over a network. The link layer manages node-to-node communication, such as within a grid computing environment. The connection layer 304 can detect and correct errors (for example, transmission errors in the physical layer 302 ). The connection layer 304 may also include a Media Access Control (MAC) layer and Logical Link Control (LLC) layer.

The network layer 306 defines the protocol for routing within a network. In other words, the network layer coordinates the transmission of data over nodes in a same network (eg, a grid computing environment). The network layer 306 can also define the processes used to structure local addressing within the network.

The transport layer 308 can manage the transmission of data and the quality of the transmission and / or reception of this data. The transport layer 308 may provide a protocol for transmitting data, such as Transmission Control Protocol (TCP). The transport layer 308 can assemble and disassemble data frames for transfer. The transport layer can also detect transmission errors that occur in the underlying layers.

The session layer 310 Can establish, maintain and manage communication links between devices in a network. In other words, the session layer controls the dialogues or the type of Communication between network devices in the network. The session layer may also define checkpoint execution, adjournment, termination, and restart procedures.

The presentation layer 312 can provide a translation for communication between the application and network layers. In other words, this layer may encrypt, decrypt, and / or format data based on data types and / or encodings that are known to be accepted by an application or network layer.

The application layer 314 interacts directly with software applications and end users and manages the communication between them. The application layer 314 can identify goals, states or availability of local resources and / or communication content or formatting using the applications.

The intra-network connection components 322 and 324 are shown to work in lower levels, such as in the physical layer, respectively 302 and connecting layer 304 , For example, a hub can work in the physical layer, a switch can work in the physical layer, and a router can work in the network layer. Inter-network connection components 326 and 328 are shown to work at higher levels, such as shifts 306 - 314 , For example, routers can work in the network layer and network devices can work in the transport, session, presentation, and application layers.

As mentioned earlier, a computing environment 314 in various embodiments, interact with and / or work with one, more, all, or any of the various layers. For example, the computing environment 314 interact with a hub (eg, via the link layer) to adjust which devices the hub communicates with. The physical layer can be served by the link layer so that it can implement such data from the link layer. For example, the computing environment 314 control from which devices it will receive data. If the computing environment 314 For example, knowing that a particular network device is turned off, defective, or otherwise unavailable or unreliable can cause the computing environment 314 instruct the hub to prevent any data from this network device from being sent to the computing environment 314 be transmitted. Such a process may be advantageous to avoid receiving data that is inaccurate or that has been affected by an uncontrolled environment. As another example, the computing environment 314 communicate with a bridge, a switch, a router or a gateway and influence which device within the system (eg system 200 ) selects the component as the destination. In some embodiments, the computing environment may 314 Interact with different layers by exchanging communication by routing or modifying existing communication with devices working on a particular layer. In another embodiment, such as. For example, in a grid computing environment, a node may determine how data is to be forwarded within the environment (eg, which node should receive certain data) based on certain parameters or information provided by other layers within the model.

As mentioned earlier, the computing environment 314 be part of a communication grid environment whose communication is as in the log 3 shown, can be implemented. For example, with reference to 2 , one or more of the machines 220 and 240 Be part of a communication grid computing environment. A grid computing environment may be used in a distributed system with non-interactive workloads where data is stored in memory in the machines or compute nodes. In such an environment, instead of a database management system, the analytical code controls the processing performed by the nodes. The data is shared by pre-distributing to the grid nodes, and the analytic code on each node loads the local data into memory. Each node may be assigned a particular task, such as part of a processing project, or to organize or control other nodes within the grid.

4 illustrates a communication grid computing system 400 comprising a plurality of control and worker nodes according to embodiments of the present technology. The communication grid computing system 400 includes three control nodes and one or more worker nodes. The communication grid computing system 400 includes the control nodes 402 . 404 and 406 , The control nodes are via the communication paths 451 . 453 and 455 communicatively connected. Therefore, the control nodes can transmit information (eg related to the communication grid or notifications) to each other and Receive information from each other. Although the communication grid computing system 400 in 4 is shown with three control nodes, the communication grid may include more or fewer than three control nodes.

The communication grid computing system (or just "communication grid") 400 also includes one or more worker nodes. In 4 are six working knots 410 - 420 shown. Even though 4 6 shows worker nodes, a communication grid according to embodiments of the present technology may include more or fewer than six worker nodes. The number of worker nodes included in a communications grid may depend, among other things, on the size of the project or data set being processed by the communications grid, the capacity of each worker node, the time allocated for the job Communication grid is designated to complete the project. Each worker node within the communication grid 400 can (wired or wireless and directly or indirectly) with the control nodes 402 - 406 be connected. Therefore, each worker node may receive information from the control nodes (eg, an instruction to perform work on a project) and may transmit information to the control nodes (eg, a result of work performed on a project). Further, worker nodes (either directly or indirectly) can communicate with each other. For example, worker nodes may transmit data to each other relating to a job being performed or a single task within a job performed by that worker node. However, in certain embodiments, worker nodes may not be connected to certain other worker nodes (communicatively or otherwise), for example. In one embodiment, worker nodes may only be able to communicate with the control node that controls them, and may not be able to communicate with other worker nodes in the communication grid, whether or not they are other worker nodes that pass through the communication node Be controlled control node that controls the worker node, or worker nodes that are controlled by other control nodes in the communication grid.

A control node may connect to an external device with which the control node can communicate (eg, a lattice user, such as a server or computer, may connect to a controller of the lattice). For example, a server or computer may connect to control nodes and transmit a project or job to the node. The project can include a dataset. The record can be of any size. Once the control node receives such a large data set project, the control node may distribute the data set or projects related to the data set to be performed by worker nodes. Alternatively, the record may be received or stored for a project with a large record from a machine other than a control node (eg, a Hadoop data node using the Hadoop Distributed File System or HDFS).

Control nodes may, among other things, maintain knowledge about the status of the nodes in the grid (i.e., grid status information), accept work requests from clients, split the work over worker nodes, coordinate the worker nodes. Worker nodes may accept work requests from a control node and provide the control node with results of the work performed by the worker node. A grid can be started from a single node (for example, a machine, a computer, a server, and so on). This first node may be assigned or may start as the primary control node that will control any additional nodes that enter the grid.

When a project is submitted for execution (eg, by a client or controller of the grid), it can be assigned to a set of nodes. After the nodes are assigned to a project, a data structure (i.e., a communicator) can be generated. The communicator can be used by the project for information to be shared between the project code running on each node. A communication handle can be generated on each node. For example, a handle is a reference to the communicator that is valid within a single process on a single node, and the handle can be used when communication between nodes is requested.

A control node, such as the control node 402 , may be referred to as the primary control node. A server, computer, or other external device may connect to the primary control node. Once the control node receives a project, the primary control node can distribute parts of the project to its worker nodes for execution. For example, if a project is on the communication grid 400 is initiated, the primary control node controls 402 the work to be performed on the project to complete the project as requested or instructed. The primary control node may distribute work to the worker nodes based on various factors, such as which ones Subsets or parts of projects can be completed most efficiently and in the correct time. For example, a worker node may perform an analysis on a portion of data that is already local to (for example, stored on) the worker node. The primary control node also coordinates and processes the results of the work performed by each worker node after each worker node completes and completes its job. For example, the primary control node may receive a result from one or more worker nodes, and the control node may organize (eg, collect and assemble) the received results and compile them to produce a complete result for the project received from the end user.

All remaining control nodes, such as the control nodes 404 and 406 , can be assigned as a backup control node for the project. In one embodiment, the backup control nodes may not control a portion of the project. Instead, backup control nodes may serve as a backup for the primary control node and take over as a primary control node if the primary control node were to fail. If a communication grid included only a single control node and the control node failed (eg, the control node is turned off or breaks), then the communication grid as a whole may fail and a project or job running on the communication grid may fail. may fail and may not complete. Although the project can be run again, such an error can cause a delay (heavy delay in some cases, such as an overnight delay) at the conclusion of the project. Therefore, a grid with multiple control nodes, including a backup control node, may be advantageous.

For example, to add another node or machine to the grid, the primary control node may open a pair of listening sockets. One socket can be used to accept client work requests, and the second socket can be used to accept connections from other grid nodes. The primary control node may be provided with a list of other nodes (eg, other machines, computers, servers) that will participate in the grid and the role that each node in the grid will fill. When the primary control node starts (for example, the first node in the grid), the primary control node can use a network protocol to start the server process on every other node in the grid. For example, command line parameters may inform each node of one or more pieces of information, including, but not limited to, the role that the node in the grid will have, the host name of the primary control node, the port number on which the primary control node accepts connections from peer nodes. The information may also be provided in a configuration file, transmitted via a secure shell tunnel, restored by a configuration server, among others. Although the other machines in the grid initially can not know the configuration of the grid, this information can also be sent through the primary control node to another node. Updates to the grid information may also be sent subsequently to these nodes.

For any control node other than the primary control node added to the grid, the control node can open three sockets. The first socket can accept work requests from clients, the second socket can accept connections from other grid elements, and the third socket can connect to the primary control node (eg, permanent). When a control node (eg, a primary control node) receives a connection from another control node, it first checks to see if the peer node in the list of configured nodes is in the grid. If it is not in the list, the control node can delete the connection. If he is in the list, he can then try to authenticate the connection. If the authentication is successful, the authentication node may send information to its peer, such as the port number on which a node listens for connections, the host name of the node, information about node authentication, and other information. When a node, such as the new control node, receives information about another active node, it is checked if it already has a connection to that other node. If he has no connection to this node, he can then connect to this control node.

Each worker node added to the grid can connect to the primary control node and any other control node on the grid. After establishing the connection, he can authenticate himself to the grid (e.g., to any control nodes, including both primary and backups, or a server or user controlling the grid). After successful authentication, the worker node can accept configuration information from the control node.

When a node connects to a communication grid (e.g., when the node is powered up or is connected to an existing node on the grid or both), the node becomes (for example, by an operating system of the grid) assigned a universally unique identifier (UUID). This unique identifier may help other nodes and external entities (devices, users, etc.) identify the node and distinguish it from other nodes. When a node is connected to the grid, the node can share its unique identifier with the other nodes in the grid. Since each node can share its unique identifier, each node can know the unique identifier of every other node on the grid. Unique identifiers may also designate a hierarchy of each of the nodes (eg, backup control nodes) within the grid. For example, the unique identifiers of each of the backup control nodes may be stored in a list of backup control nodes to indicate an order in which the backup control nodes assume for a failed primary control node to become a new primary control node. However, a hierarchy of nodes may also be determined using methods other than using the unique identifiers of the nodes. For example, the hierarchy may be predetermined or assigned based on other predetermined factors.

The grid can add new machines at any time (for example, initiated by any control node). After adding a new node to the grid, the control node may first add the new node to its table of grid nodes. The control node may then notify any other control node of the new node. The nodes that receive the notification may confirm that they have updated their configuration information.

The primary control node 402 may, for example, one or more communications to the backup control nodes 404 and 406 (and, for example, to other control or worker nodes within the communication grid). Such communications may be sent periodically at fixed time intervals between known fixed stages of execution of the project, among other protocols. The from the primary control node 402 transmitted communications may be of various types and may include a variety of types of information. For example, the primary control node 402 Snapshots (such as status information) of the communication grid are transmitted, leaving the backup control node 404 always has a last snapshot of the communication grid. For example, the snapshot or grid state may include the structure of the grid (including, for example, the worker nodes in the grid, the unique identifiers of the nodes or their relationships with the primary control node) and the status of a project (eg, the status of the part of the project from each worker node ). The snapshot may also include analyzes or results received from worker nodes in the communication grid. The backup control nodes can receive and store the backup data received from the primary control node. The backup control nodes may transmit a request for such a snapshot (or other information) from the primary control node, or the primary control node may periodically send such information to the backup control nodes.

As mentioned previously, if the primary control node fails, the backup data may allow the backup control node to act as the primary control node without the grid having to restart the project from the beginning. If the primary control node fails, the backup control node, which will take over as the primary control node, may retrieve the most recent version of the snapshot received from the primary control node and use the snapshot to continue the project from the stage of the project through the backup data is shown. This can prevent a failure of the project as a whole.

A backup control node may use various methods to determine that the primary control node has failed. In one example of such a method, the primary control node may send communication (eg, periodically) to the backup control node indicating that the primary control node is functioning and not failed, such as heartbeat communication. The backup control node may determine that the primary control node has failed if the backup control node has not received heartbeat communication for a certain predetermined period of time. Alternatively, a backup control node may also receive communication from the primary control node itself (before it failed) or from a worker node that the primary control node has failed, for example because the primary control node has not communicated with the worker node.

Various methods may be performed to determine which backup control node of a set of backup control nodes (eg, backup control nodes 404 and 406 ) for the failed primary control node 402 will take over and become the new primary control node. For example, the new primary control node may be selected based on a ranking or "hierarchy" of backup control nodes based on their unique identifiers. In an alternative embodiment For example, a backup control node may be assigned by another device in the communication grid or by an external device (eg, a system infrastructure or an end user, such as a server or computer that controls the communication grid) as the new primary control node. In another alternative embodiment, the backup control node acting as the new primary control node may be designated based on bandwidth or other statistics over the communication grid.

A worker node within the communication grid may also fail. If a worker node fails, the work performed by the failed worker node can be redistributed among the worker operable nodes. In an alternative embodiment, the primary control node may transmit a communication to each of the operational worker nodes still on the communication grid, that each of the worker nodes should also deliberately fail. After all worker nodes fail, they can each retrieve the last saved checkpoint of their status and restart the project from that checkpoint to minimize the lost progress of the running project.

5 FIG. 12 illustrates a flowchart depicting an example process for adjusting a communication grid or work project in a communication grid after a node has failed, in accordance with embodiments of the present technology. FIG. For example, the process may include receiving grid status information including a project status of a portion of a project being executed by a node in the communications grid, as in the process 502 described. For example, a control node (eg, a backup control node connected to a primary control node and a worker node on a communication grid) may receive grid status information, the grid status information including a project status of the primary control node or a project status of the worker node. The project status of the primary control node and the project status of the worker node may include a status of one or more parts of a project being executed by the primary nodes and worker nodes in the communication grid. The process may also include storing the grid status information as in the process 504 described. For example, a control node (eg, a backup control node) may locally store the received grid status information within the control node. Alternatively, the grid status information may be sent to another device for storage where the control node may have access to the information.

The process can be in the process 506 also include receiving an error communication corresponding to a node in the communication grid. For example, a node may receive an error communication that includes an indication that the primary control node has failed, prompting a backup control node to take over for the primary control node. In an alternative embodiment, a node may receive an error that a worker node has failed, thereby prompting a control node to reassign the work performed by the worker node. The process may also involve reassigning a node or part of the project being executed by the failed node, as in the process 508 described. For example, a control node may designate the backup control node as a new primary control node based on the error communication after receiving the error communication. If the failed node is a worker node, a control node may identify a project status of the failed worker node using the snapshot of the communications grid, the project status of the failed worker node including a status of a portion of the portion of the project being executed by the failed worker node at downtime.

The process may also include receiving updated grid status information based on the reallocation as in the process 510 and transmitting a set of instructions based on the updated grid status information to one or more nodes in the communication grid, as in the process 512 described include. The updated grid status information may include an updated project status of the primary control node or an updated project status of the worker node. The updated information may be transmitted to the other nodes in the grid to update their expired stored information.

6 illustrates part of a communication grid computing system 600 comprising a control node and a worker node according to embodiments of the present technology. The communication grid computing system 600 for purposes of illustration includes a control node (control node 602 ) and a worker node (worker node 610 ), but may also include more worker and / or control nodes. The control node 602 is via the communication path 650 communicatively with the worker node 610 connected. Therefore, the control node 602 Information (eg related to the Communication grid or notifications) via the path 650 at the worker node 610 transmit and receive information from him.

Similar to in 4 includes the communication grid computing system (or just "communication grid") 600 Data processing node (control node 602 and worker nodes 610 ). The knots 602 and 610 include multi-core data processors. Every node 602 and 610 includes a Grid-Enabled Software Component (GESC) 620 which is executed on the data node associated with this node and via an interface with the buffer memory 622 which is also associated with this node is connected. Every node 602 and 610 includes database management software (DBMS) 628 stored on a database server (not shown) at the control node 602 and on a database server (not shown) at the worker node 610 is performed.

Each node also includes a data store 624 , The data storage 624 , similar to the data memories connected to the network 110 in 1 and data stores 235 in 2 , are used to store data to be processed by the nodes in the computing environment. The data storage 624 may also store any intermediate or final data generated by the computing system after it has been processed, for example, in non-volatile memory. However, in certain embodiments, the configuration of the grid computing environment allows its operations to be performed so that intermediate and final data results can be stored only in volatile memory (eg, RAM) without storing intermediate or final data results in nonvolatile types of memory have to. Storing such data in volatile memory may be useful in certain situations, such as when the grid receives queries (eg, ad hoc) from a client, and when responses generated by processing large amounts of data are promptly or on-the-fly. the-fly must be generated. In such a situation, the grid may be designed to hold the data within the memory so that responses can be generated at various levels of detail and so that a client can interrogate that information interactively.

Each node also includes a user-defined function (UDF) 626 , The UDF provides a mechanism for the DBMS 628 ready to transfer data to or from the database stored in the data stores 624 stored, which are managed by the DBMS. For example, the UDF 626 are called by the DBMS to provide data to the GESC for processing. The UDF 626 can establish a socket connection (not shown) with the GESC to transfer the data. Alternatively, the UDF 626 Transfer data to the GESC by writing data to shared memory that is accessible to both the UDF and the GESC.

The GESC 620 at the node 602 and 620 can be over a network, such as the one in 1 shown network 108 be connected. Therefore, the nodes can 602 and 620 communicate over the network using a predetermined communication protocol, such as the Message Passing Interface (MPI). Every GESC 620 may participate in a point-to-point communication with the GESC at another node or in collective communication with multiple GESCs over the network. The GESC 620 Each node may contain identical (or nearly identical) software instructions. Each node may be able to operate as either a control node or a worker node. The GESC at the control node 602 can via a communication path 652 with a client device 630 communicate. In particular, the control node 602 with the client application 632 communicate by the client device 630 is hosted to receive queries and respond to those queries after large amounts of data have been processed.

The DBMS 628 may be the creation, maintenance and use of the database or data structure (not shown) within a node 602 or 610 Taxes. The database can be stored in the datastores 624 organize stored data. The DBMS 628 at the control node 602 can accept requests for data and transfer the corresponding data for the request. In such a process, data collections can be distributed across multiple physical locations. In this example, each node stores 602 and 610 a portion of the total data managed by the management system in its associated data store 624 ,

Further, the DBMS may be responsible for data loss protection using replication techniques. The replication involves providing a backup copy of data stored on a node or on one or more other nodes. Therefore, if one node fails, data from the failed node can be recovered from a replicated copy located at another node. However, as discussed herein with reference to 4 described, data or Status information for each node in the communication grid can also be shared with each node on the grid.

7 FIG. 12 illustrates a flowchart depicting an exemplary method for executing a project within a grid computing system, in accordance with embodiments of the present technology. As with reference to 6 At the control node, the GESC may display data with a client device (eg, client device 630 ) to receive queries to run a project and respond to those queries after large amounts of data have been processed. The query may be transmitted to the control node, where the query may include a request to execute a project, as in the process 702 described. The query may contain instructions about the type of data analysis to be performed in the project, and whether the project should be executed using the grid-based computing environment, as in the process 704 shown.

To initiate the project, the control node can determine whether the query requests the use of the grid-based computing environment to run the project. If the determination is no, the control node initiates the execution of the project in a solo environment (eg at the control node) as in the process 710 described. If the determination is yes, the control node may initiate the execution of the project in the grid-based computing environment as in the process 706 described. In such a situation, the request may include a requested configuration of the grid. For example, the request may include a number of control nodes and a number of worker nodes to be used in the grid when the project is executed. After the project has completed, the control node can transmit the results of the analysis provided by the grid as in the process 708 described. Whether the project is running in a solo or grid-based environment, the control node provides the results of the project.

As with reference to 2 As mentioned earlier, the computing environments described herein may collect data (eg, as received from network devices, such as sensors, such as network devices 204 - 209 in 2 and client devices or other sources) to be processed as part of a data analytics project, and data can be received in real time as part of a streaming analytics environment (e.g., ESP). The data may be collected using a variety of sources as communicated across different types of networks or locally, such as on a real-time streaming basis. For example, network devices may periodically receive data from network device sensors as the sensors continuously detect, monitor, and track changes in their environments. In particular, an increasing number of distributed applications develop or generate continuously flowing data from distributed sources by applying queries to the data before distributing the data to geographically dispersed recipients. An event stream processing engine (ESPE) can continuously apply the queries to the data as it is received and determines which entities should receive the data. Client or other devices may also subscribe to the ESPE or other devices that process ESP data so that they may receive data after processing based on, for example, the entities designated by the processing engine. For example, the client devices 230 in 2 the ESPE in the computing environment 214 subscribe to. In another example, the event subscribing devices may 874a -C, referring to 10 be subscribed to, also subscribe to the ESPE. The ESPE may determine or define how input data or event streams from network devices or other publishers (eg, network devices 204 - 209 in 2 ) are converted into meaningful output data from subscribers, such as the client devices 230 in 2 to be consumed.

8th FIG. 12 illustrates a block diagram including components of an Event Stream Processing Engine (ESPE) according to embodiments of the present technology. The ESPE 800 can one or more projects 802 include. A project can be considered a second level container in one of the ESPE 800 managed engine model, where a thread pool size for the project can be defined by a user. Each project of one or more projects 802 can be one or more continuous queries 804 include data streams that are data transformations of incoming event streams. The one or more continuous queries 804 can have one or more source windows 806 and one or more derived windows 808 include.

The ESPE may receive streaming data over a period of time related to particular events, such as events or other data captured by one or more network devices. The ESPE can perform operations associated with processing data generated by the one or more devices are generated. For example, the ESPE may receive data from the one or more network devices 204 - 209 receive that in 2 are shown. As mentioned earlier, the network devices may include sensors that sense different aspects of their environments, and may gather data over time based on these acquired observations. For example, the ESPE may be within one or more of the machines 220 and 240 , in the 2 shown are implemented. The ESPE can be implemented within such a machine by an ESP application. An ESP application can embed an ESPE with its own dedicated thread pool or thread pools within its scope, where the main application thread can perform application-specific work, and at least process the ESPE event streams into processing objects by creating an instance of a model. The engine container is the top-level container in a model that contains the resources of one or more projects 802 managed. For example, in an illustrative embodiment, only one ESPE 800 for each instance of the ESP application, and the ESPE 800 can have a unique engine name. In addition, the one or more projects 802 each have unique project names, and each query may have a unique continuous query name and a unique source window of the one or more source windows 806 kick off. The ESPE 800 can be persistent or not.

Continuous query modeling involves defining directed graphs of event stream manipulation and transformation windows. A window in the context of event stream manipulation and transformation is a processing node in an event stream processing model. A window in a continuous query can perform aggregations, calculations, pattern adjustments, and other operations on data flowing through the window. A continuous query may be described as a directed graph of source, relational, pattern matching, and procedural windows. The one or more source windows 806 and the one or more derived windows 808 represent the continuous execution of queries that generate updates to a query result set when new event blocks through the ESPE 800 stream. For example, a directed graph is a set of nodes connected by edges, the edges having a direction associated therewith.

An event object may be described as a packet of data accessible as a collection of fields, where at least one of the fields is defined as a key or unique identifier (ID). The event object can be created using a variety of formats, including binary, alphanumeric, XML, and so on. Each event object may include one or more fields designated as the primary identifier (identifier, ID) for the event, such that the ESPE 800 Operational codes (opcodes) for events, including insert, update, upsert, and delete. Upsert opcodes update the event if the key field already exists; otherwise the event will be inserted. By way of illustration, an event object may be a packed binary representation of a set of field values and may include both metadata and field data associated with an event. The metadata may include an opcode indicating whether the event represents an insert, update, delete, or upsert, a set of flags indicating whether the event is a normal, partial, or retention-generated event from the retention policy management and a set of microsecond timestamps that can be used for latency measurements.

An event block object can be described as a grouping or package of event objects. An event stream can be described as a flow of event block objects. A continuous query of the one or more continuous queries 804 converts a source event stream that consists of streaming event block objects that are in the ESPE 800 in one or more output event streams using the one or more source windows 806 and one or more derived windows 808 around. A continuous query can also be considered as data flow modeling.

The one or more source windows 806 are located at the top of the directed graph and have no windows that are inserted into them. Event streams are in the one or more source windows 806 and from there the event streams may be directed to the next set of connected windows as defined by the directed graph. The one or more derived windows 808 are all instantiated windows that are not source windows and have other windows that stream events into them. The one or more derived windows 808 can perform calculations or transformations on the incoming event streams. The one or more derived windows 808 convert event streams based on the type of window (ie operators such as join, filter, calculate, aggregate, copy, pattern match, procedural, union, etc.) and window settings. If event streams in the ESPE 800 they will be released queried continuously, and the resulting sets of derived windows in these queries are continually updated.

9 FIG. 12 illustrates a flowchart depicting an example process including operations performed by an event stream processing engine, in accordance with some embodiments of the present technology. As already mentioned, the ESPE defines 800 (or an associated ESP application) how input event streams are transformed into meaningful output event streams. In particular, the ESP application may define how to convert input event streams from publishers (eg, network devices that provide captured data) into meaningful output event streams that are consumed by subscribers (eg, a Data Analytics project run by a machine or a set of machines).

Within the application, a user may interact with one or more user interface windows that are presented to the user in a display under the control of ESPE, independently or through a browser application in a user-selectable order. For example, a user may execute an ESP application that causes the presentation of a first user interface window, which may include a plurality of menus and selectors, such as drop-down menus, buttons, text boxes, hyperlinks, etc., associated with the ESP application, as understood by those skilled in the art. As will be understood by those skilled in the art, various operations may be performed in parallel, for example, using a plurality of threads.

In the process 900 For example, an ESP application may define and launch an ESPE, thereby causing an ESPE on a device, such as the machine 220 and or 240 , is instantiated. In one operation 902 the engine container is created. By way of illustration, the ESPE 800 is instantiated using a function call that specifies the engine container as a manager for the model.

In an operation 904 become the one or more continuous queries 804 through the ESPE 800 instantiated as a model. The one or more continuous queries 804 can be instantiated with a dedicated thread pool or thread pools that generate updates when new events occur through the ESPE 800 stream. By way of illustration, the one or more continuous queries 804 generated to the business processing logic within the ESPE 800 to model events within the ESPE 800 predict a physical system within the ESPE 800 to model the physical state of the system within the ESPE 800 For example, as already mentioned, the ESPE 800 can be used to assist sensor data monitoring and management (eg, the detection force, torque, load, strain, position, temperature, air pressure, fluid flow, chemical properties, resistance, electromagnetic fields, radiation, irradiation, proximity, acoustics , Humidity, distance, speed, vibration, acceleration, electric potential or electric current).

The ESPE 800 can analyze and process events in motion or "event streams". Instead of storing data and retrieving the stored data, the ESPE 800 Save queries and stream data through them to allow continuous analysis of the data as it is received. The one or more source windows 806 and the one or more derived windows 808 may be generated based on the relational, pattern matching, and procedural algorithms that transform the input event streams into the output event streams to model, simulate, evaluate, test, based on the defined continuous query model and application on the streamed data to predict, etc.

In one operation 906 becomes a publish / subscribe (pub / sub) capability for the ESPE 800 initialized. In an illustrative embodiment, a pub / sub capability will be for each project of the one or more projects 802 initialized. To the pub / sub-ability for the ESPE 800 initialize and activate, a port number can be provided. Pub / sub clients can use a hostname of an ESP device that executes the ESPE and the port number to provide pub / sub connections to the ESPE 800 manufacture.

10 illustrates an ESP system 850 that has an interface between the publishing device 872 and the event subscribing devices 874a -C, according to embodiments of the present technology. The ESP system 850 may be the ESP device or the ESP subsystem 851 , the event publishing device 872 , an event-subscribing device A. 874a , an event-subscribing device B 874b and an event subscribing device C 874c include. Input event streams are sent to the ESP device 851 through the publishing device 872 output. In alternative embodiments, the input event streams may be generated by a plurality of publishing devices. The plurality of publishing devices may also issue event streams to other ESP devices. The one or more continuous queries made by the ESPE 800 can be instantiated and parse the input event streams to form output event streams sent to the event subscribing device A. 874a , the event subscribing device B 874b and the event subscribing device C 874c be issued. The ESP system 850 may include a larger or smaller number of event subscribing devices from event subscribing devices.

Publish-Subscribe is a message-oriented interaction paradigm based on indirect addressing. Processed data recipients give their interest in receiving information from the ESPE 800 by subscribing to certain classes of events while sources of information send events to the ESPE 800 without directly addressing the receiving parties. The ESPE 800 coordinates the interactions and processes the data. In some cases, the data source receives confirmation that the issued information has been received by a data receiver.

A publish / subscribe API can be described as a library that allows an event publisher, such as the publishing device 872 , allows event streams into the ESPE 800 or an event subscriber, such as the event subscribing device A. 874a , the Event Subscribing Device B 874b and the event subscribing device C 874c , allows event streams from the ESPE 800 to subscribe. As an illustration, one or more publish / subscribe APIs can be defined. Using the publish / subscribe API, an event publishing application can use event streams in a running event stream processor project source window from the ESPE 800 Issue the Event Subscription application and an Event Stream Processor project source window from the ESPE 800 subscribe to.

The publish / subscribe API provides cross-platform connectivity and endianness compatibility between the ESP application and other networked applications, such as event publishing applications running on the publishing device 872 to be instantiated, and event subscription applications attached to one or more of the event subscribing device A 874a , the Event Subscribing Device B 874b and the event subscribing device C 874c to be instantiated.

With reference to 9 initializes the process 906 the publish / subscribe capability of the ESPE 800 , In one operation 908 become one or more projects 802 started. The one or more started projects may be run in the background on an ESP device. In one operation 910 becomes an event block object of one or more event publishing device computing devices 872 receive.

The ESP subsystem 800 can be a publishing client 852 , the ESPE 800 , a Subscribing Client A 854 , a subscribing client B 856 and a subscribing client C 858 include. The publishing client 852 can be started by an event publishing application used by the publishing device 872 using the publish / subscribe API. The Subscribing Client A 854 can be started by an event subscription application A, which in the event subscribing device A 874a using the publish / subscribe API. The subscribing client B 856 can be started by an event subscription application B, which in the event subscribing device B 874b using the publish / subscribe API. The subscribing client C 858 can be started by an event subscription application C, which in the event subscribing device C 874c using the publish / subscribe API.

An event block object containing one or more event objects is placed in a source window of the one or more source windows 806 from an instance of an event publishing application on the event publishing device 872 injected. For example, the event block object can be generated by the event publishing application and can be used by the publishing client 852 be received. A unique ID can be maintained if the event block object is between the one or more source windows 806 and / or the one or more derived windows 808 the ESPE 800 and the subscribing client A 854 , Subscribing client B 806 and subscribing client C 808 and to the event subscription device A. 874a , to the event subscription device B 874b and the event subscription device C 874c is handed over. The publishing client 852 may also generate and include a unique embedded transaction ID in the event block object if the event block object is replaced by a continuous Query is processed, as well as the unique ID that the publishing device has 872 assigned to the event block object.

In one operation 912 The event block object is passed through the one or more continuous queries 804 processed. In one operation 914 the processed event block object is sent to one or more computing devices of the event subscribing devices 874a -C output. For example, the subscribing client A 804 , the subscribing client B 806 and the subscribing client C 808 the received event block object respectively to the event subscription device A. 874a , the event subscription device B 874b and the event subscription device C 874c send.

The ESPE 800 maintains the event block containment aspect of the received event blocks from when the event block is issued to a source window, and works through the directed graph that passes through the one or more continuous queries 804 is defined with the various event translations prior to output to Subscriber. Subscribers may correlate a set of subscribed events with a group of issued events by comparing the unique ID of the event block object that a publisher, such as the publishing device, is comparing 872 to which the event block object with the event block ID received from the subscriber has attached.

In one operation 916 a determination is made as to whether the processing is stopped or not. If processing is not stopped, processing moves in the process 910 to further receive the one or more event streams containing event block objects from, for example, one or more network devices. If processing is stopped, processing proceeds in one operation 918 continued. In the process 918 the started projects are stopped. In the process 920 the ESPE is switched off.

As previously mentioned, in some embodiments, large data is processed for an analysis project after the data is received and stored. In other embodiments, distributed applications process continuously flowing data in real time from distributed sources by applying queries to the data before distributing the data to geographically dispersed recipients. As mentioned earlier, an Event Stream Processing Engine (ESPE) can continuously apply the queries to the data as it is received and determines which entities receive the processed data. This allows large amounts of data to be received and / or collected in a variety of environments for real-time processing and distribution. For example, data as described with reference to 2 can be collected from network devices that may include devices within the Internet of Things, such as devices within a home automation network. However, such data can be collected from a variety of different resources in a variety of different environments. In such a situation, embodiments of the present technology enable real-time processing of such data.

Aspects of the present disclosure provide technical solutions to technical problems, such as computational problems that occur when an ESP device fails, resulting in a complete service interruption and potentially significant data loss. Data loss can be catastrophic if the streamed data supports mission-critical operations, such as those that support an ongoing manufacturing or drilling operation. One embodiment of an ESP system achieves rapid and seamless failover of the ESPE that is performed on the plurality of ESP devices without service interruption or data loss, thereby significantly improving the reliability of an operational system based on live or real-time processing of the data streams becomes. The event publishing systems, the event subscribing systems, and any ESPEs that are not executed on a failed ESP device are unaware of and are not effected by the failed ESP device. The ESP system can include thousands of event publishing systems and event subscribing systems. The ESP system maintains the failover logic and awareness within the boundaries of the Out-Messaging Network Connector and the Out-Messaging Network Device.

In one embodiment, a system is provided to support failover on event stream processing (ESP) event blocks. The system includes, among other things, an out-messaging network device and a computing device. The computing device includes, among other things, a processor and a computer readable medium operably coupled to the processor. The processor is designed to execute an ESP engine (ESPE). The computer-readable medium has instructions stored thereon that, when executed by the processor, cause the computing device to assist in failover. An event block object is identified by the ESPE with a unique ID received. A first status of the computing device is determined as "active" or "standby". If the first status is "active", a second status of the computing device is determined to be "new active" or "not newly active". "New active" is determined when the computing device is switched from a standby state to an active state. When the second status is "new active", a most recently issued event block object identifier is identified which uniquely identifies a most recently issued event block object. A next event block object is selected by a non-transitory computer readable medium accessible by the computing device. The next event block object has an event block object identifier that is greater than the particular last issued event block object identifier. The selected next event block object is issued to an out-messaging network device. If the second state of the computing device is not "re-active", the received event block object is issued to the out-messaging network device. When the first state of the computing device is "standby", the received event block object is stored in the non-transitory computer-readable medium.

11A Figure 12 illustrates a block diagram of one embodiment of a distributed processing system 1000 containing one or more storage devices 1100 holding a storage device grid 1001 can form a coordinating device 1300 , several node devices 1500 that a knot device grid 1005 can form, and / or a viewing device 1800 integrated. 11B Figure 12 illustrates a block diagram of an alternative embodiment of the distributed processing system 1000 in which an alternative embodiment of the node devices 1500 Features of the one or more storage devices 1100 integrates and / or performs functions thereof. In both of these embodiments of the distributed processing system 1000 and as explained in more detail, the node devices 1500 together as the grid 1005 under the control of the coordinator 1300 are operated, wherein each of the plurality of node devices 1500 the same task at least partially parallel to another of several record partitions 1131 a record 1130 performs on the multiple node devices 1500 are distributed.

As shown, these devices can 1100 . 1300 . 1500 and 1800 Communications in between in connection with the assignment and execution of tasks of an analysis routine 1210 with one or more records 1130 change. Such communications can exchange the node status 1535 , Record partitions 1131 and / or metadata 1135 a record 1130 , the analysis routine 1210 and / or task routines 1211 of it, CPU task routines 1571 , GPU Task Routines 1671 and / or result data 1830 lock in. However, one or more of the devices 1100 . 1300 . 1500 and or 1800 also over the net 1999 exchange other data that is in no way related to any assignment or performance of tasks of any analysis routine. In various embodiments, the network may 1999 be a single network that may extend within a single building or other relatively limited area, be a combination of connected networks that may extend a considerable distance, and / or include the Internet. Thus, the network can 1999 based on any variety (or combination) of communication technologies through which communications can be effected, including, without limitation, wired technologies using electrically and / or optically conductive cabling, and wireless technologies including infrared, radio frequency (RF) - or use other forms of wireless transmission.

With reference to 11A For example, in various embodiments, each of the one or more memory devices 1100 one or more processors 1150 , a store 1160 and a network interface 1190 integrate with each of the one or more storage devices 1100 with the net 1999 to pair. The memory 1160 can be a control routine 1140 , one or more analysis routines 1210 , each one or more task routines 1211 can integrate one or more records 1330 , each metadata 1135 integrate and / or one or more record partitions 1131 one or more records 1130 to save. The control routine 1140 can integrate a sequence of instructions on the processor 1150 from each of the one or more storage devices 1100 are operable to implement logic to perform various functions. The processor 1150 from each of the storage devices 1100 can the network interface 1190 operate to the analysis routine 1210 and / or one or more of the task routines 1211 of which with the coordinator 1300 exchange. Alternatively or additionally, the processor 1150 from each of the storage devices, the network interface 1190 operate to exchange one or more record partitions 1131 with one or more of the node devices 1500 over the net 1999 with the coordinator 1300 to coordinate and to effect this exchange. In embodiments in which more of the memory devices 1100 together as the storage device grid 1001 can be operated, the sequence of instructions of the control routine 1140 on the processor 1150 from each one of these storage devices 1100 be operable to perform various functions at least partially parallel to the processors 1150 from others of the storage devices 1100 perform.

In some embodiments, the processors may 1150 the storage devices 1100 work together to perform a collection function in which each of the processors 1150 a corresponding one of the network interfaces 1190 operates to data elements of one or more of the records 1130 over the net 1999 and can transfer the data items into the one or more records over a period of time 1130 assemble. In such embodiments, data elements of a data set 1130 over the net 1999 and / or otherwise received by one or more other devices (not shown). By way of example, a plurality of remote sensor devices (eg, geological sensors distributed over a particular geological region or particle detection sensors disposed on different parts of a particle accelerator) may generate numerous data elements which then travel over the network 1999 to the storage devices 1100 be provided where the numerous data elements are then assembled to a record 1130 train. In other embodiments, the memory devices 1100 one or more of the records 1130 a plurality of other devices (not shown), such as a grid of other node devices. By way of example, such other devices may perform one or more processing operations that include a record 1130 (For example, use a Bayesian analysis to derive a prediction of the behavior of humans in a simulation of evacuating a burning building or to derive a prediction of the behavior of structural components of a bridge in response to different wind currents), and then make a record 1130 as an output to the memory device grid 1001 transfer.

Each of the one or more records 1130 may include any of a variety of types of data associated with any of a variety of items. By way of example, each of the records 1130 Include scientific observation data relating to geological and / or meteorological events or sensors used in laboratory experiments in fields such as chemistry or physics. In another example, the dataset may include displaying activities performed by a random sample of individuals of a population of humans in a selected country or community, or a population of endangered species being studied in the wild. As shown, each of the records can 1130 metadata 1135 Incorporate the ads to provide structural features, including, but not limited to, aspects of the way in which data items are organized and / or within each record 1130 be made accessible.

The tasks with respect to which the task routines 1211 the analysis routine 1210 may cause one or more processors to perform any of a variety of data analysis tasks, data transformation tasks, and / or data normalization tasks. The data analysis tasks may include, but are not limited to, searches and / or statistical analyzes involving the derivation of approximations, numerical characterizations, models, hypothesis evaluations, and / or predictions (e.g., a prediction by a Bayesian analysis of actions of a crowd, trying to escape from a burning building or the behavior of bridge components in response to wind power). The data transformation tasks may include, but are not limited to, sorting, row and / or column based mathematical operations, row and / or column based filtering using one or more data elements of a row or column, and / or reordering data items within a data object. The data normalization tasks may include normalization times of the day, dates, monetary values (eg, normalization to a single currency unit), character spacing, use of delimiters (eg, normalizing the use of points and commas in numeric values), use of formatting codes, Use of big or little endian encoding, use or lack of character bits, sets of bits used to represent integers and / or floating point values (eg, bytes, words, double words or quadwords), etc. lock in.

In various embodiments, the coordinating device 1300 one or more of a processor 1350 , a store 1360 , an input device 1320 , an ad 1380 and a network interface 1390 integrate to the coordinating device 1300 with the net 1999 to pair. The memory 1360 can be a control routine 1340 , the metadata 1135 a record 1130 , the analysis routine 1210 , Node status 1535 , Assignment data 1330 , Task delay data 1335 , a compile routine 1440 , Compile rules 1434 , the CPU task routines 1571 , the GPU task routines 1671 and / or the result data 1830 to save. The control routine 1340 and the compile routine 1440 (if within the memory 1360 present) can each integrate a sequence of instructions on the processor 1350 operable to implement logic to perform various functions.

In various embodiments, each of the node devices 1500 one or more of a CPU 1550 , a store 1560 and a network interface 1590 integrate to each of the node devices 1500 with the net 1999 to pair. The memory 1560 can be a control routine 1540 , one or more record partitions 1131 , an instance of the node status 1535 , the compiler routine 1440 , the compilation rules 1434 , one or more of the task routines 1211 and / or one or more of the CPU task routines 1571 to save. The control routine 1540 and the compile routine 1440 (if inside the memory 1560 present) can integrate a sequence of instructions on the CPU 1550 from each of the node devices 1500 operable to implement logic to perform various functions. In embodiments in which a plurality of the node devices 1500 together as the node device grid 1005 can be operated, the sequence of instructions of the control routine 1540 on the CPU 1550 from each of these node devices 1500 be operable to perform various functions at least partially parallel to the CPUs 1550 from other of the node devices 1500 perform.

At least a subset of the node devices 1500 can also have a graphics controller 1600 integrate one or more of a GPU 1650 and a memory 1660 can integrate. The memory 1660 can be a control routine 1640 , one or more record partitions 1131 and / or one or more of the GPU task routines 1671 to save. The control routine 1640 can integrate a sequence of statements on the GPU 1650 from each of the node devices 1600 are operational, which is the graphics controller 1600 integrated to implement logic to perform various functions. In embodiments in which a plurality of the node devices 1600 together as the node device grid 1005 can be operated, the sequence of instructions of the control routine 1640 on the GPU 1650 of the graphics controller 1600 from each of these node devices 1500 be operable to perform various functions at least partially parallel to the GPUs 1650 of the graphics controller 1600 from other of the node devices 1500 perform.

As in 12 represented, the memory can 1560 in a volatile memory 1561 and a non-volatile memory 1563 be split, and the memory 1660 can be a volatile memory 1661 but may not include non-volatile memory. The volatile memory 1561 and 1661 each with one or more volatile memory components 1562 and 1662 be implemented. The volatile memory components 1562 and 1662 Any of a variety of memory technologies may be used to provide relatively quick access to data and / or routines stored therein (eg, the illustrative record partition shown) 1131 ), but are unable to retain the data and / or routines stored therein without a continuous supply of electrical power. Such technologies include, but are not limited to, any of a variety of types of Random Access Memory (RAM). The non-volatile memory 1563 can with one or more non-volatile memory components 1564 be implemented. The one or more non-volatile memory components 1564 For example, each may employ a storage technology capable of storing data and / or routines stored therein, whether or not electrical power continues to be provided but is unable to provide access to data and / or routines, such fast is like that provided by various volatile memory technologies on which the volatile memory 1561 and or 1661 can be based. Such technologies include, but are not limited to, any of a variety of technologies employing ferromagnetic and / or optical storage media.

Due to the differences in their respective technologies, the non-volatile memory can 1563 have a much larger storage capacity than one of the volatile memory 1561 or 1661 , Thus, pages of data and / or routines within the non-volatile memory 1563 stored in each of the volatile memories 1561 and 1661 as a mechanism to the CPU 1550 and the GPU 1650 to be activated, swapped in and out, for faster access to the respective volatile memory 1561 and 1661 while overcoming the more limited storage capacities of each. In particular, and as shown, the CPU can 1550 by the execution of a page component 1541 the control routine 1540 to cause such a page swap to both its own operation and the operation of the GPU 1650 to support. The need for the CPU 1550 to use a page exchange on behalf of the GPU 1650 may result from the use of the one or more driver routines (not shown) provided by the CPU 1550 run it to the CPU 1550 to allow access to one or more non-volatile memory components 1564 access.

With reference to 11A Integrates the viewing device 1800 in various embodiments, one or more of a processor 1850 , a store 1860 , an input device 1820 , an ad 1880 and a network interface 1890 to the viewing device 1800 with the net 1999 to pair. The memory 1860 can be one or more of a control routine 1840 , the analysis routine 1210 and the result data 1830 to save. The control routine 1840 can integrate a sequence of instructions on the processor 1850 are operable to implement logic to perform various functions. The processor 1850 can through its execution of the control routine 1840 be initiated, the network interface 1890 operate the results data 1830 from one of the node devices 1500 or from the coordinator 1300 over the net 1999 after completion of the analysis routine 1210 to recieve. In some embodiments, the processor 1850 also caused a visualization based on the result data 1830 to generate a representation of the results of performance by a plurality of the node devices 1500 the tasks of the task routines 1211 the analysis routine 1210 on the display 1880 showcase.

Alternatively or additionally, the processor 1850 by its execution of the control routine 1840 be prompted the ad 1880 and / or the input device 1820 to provide a user interface by which an operator of the viewing device 1800 can provide an input therein. This input may issue a command to run the parsing routine 1210 over several of the node devices 1500 of the node device grid 1005 Include an analysis with at least one of the records 1130 performed by the one or more storage devices 1100 get saved. In response to receipt of the input command, the processor may 1850 be initiated, the network interface 1890 operate on the command and / or the analysis routine 1210 over the net 1999 to the coordinator 1300 to convey.

With reference to 11B the alternative embodiment of the distributed processing system differs 1000 Illustrated therein, from the embodiment 11A by placing the one or more storage devices 1100 not included. Instead, the node devices 1500 of the alternative embodiment 11B directly the function of storing the one or more records 1130 perform, eliminating the need for one or more storage devices 1100 the embodiment of the distributed processing system 1000 out 11A is excluded.

With reference to both embodiments of the two 11A and 11B It should be noted that in some embodiments, that of the coordinator 1300 performed functions of one of the node devices 1500 instead of the coordinator 1300 can be executed (eg, embodiments, the coordination device 1300 do not include). In such embodiments, one of the node devices 1500 additionally the metadata 1135 one of the records 1130 from one of the storage devices 1100 (or from one of the other node devices 1500 in embodiments in which the node devices 1500 the storage function of the one or more storage devices 1100 perform). In addition, such one of the node devices 1500 additionally the node status 1535 from other of the node devices 1500 receive. Further, such one of the node devices 1300 additionally the task routines 1211 , the CPU task routines 1571 and / or the GPU task routines 1671 to other of the node devices 1500 transfer.

13A FIG. 12 illustrates one embodiment of assignment of tasks and compilation of task routines performed by the coordinator 1300 in one of the embodiments of the distributed processing system 1000 from one of 11A or 11B can be performed. 13B Figure 12 illustrates an alternative embodiment of assigning tasks that may be performed by the coordinator while compiling task routines through the node devices 1500 which are selected to be assigned, the tasks in one of the embodiments of the distributed processing system 1000 from one of 11A or 11B perform.

With reference to 13A can during the execution of the control routine 1340 the processor 1350 the coordinator 1300 be prompted for metadata 1135 to receive indicative of structural features of any of the records 1130 and / or the analysis routine 1210 from the one or more storage devices 1100 and / or the viewing device 1700 are. The processor 1350 can also be caused by the node devices 1500 the node status 1535 which are indicative of the processing, storage and / or network access resources that are in each of the node devices 1500 integrated, and the extent to which each of these resources is currently available. The processor 1350 can the metadata 1135 , the analysis routine 1210 and / or the node status 1535 to initial assignments of at least one initial task of the analysis routine 1210 to selected ones of the node devices 1500 and an initial distribution of the record partitions 1131 to the selected one of the node devices 1500 derive.

When executing the compile routine 1440 can the processor 1350 the executable statements within each of the task routines 1211 the analysis routine 1210 to analyze those of the task routines 1211 to identify the embarrassingly parallel execution by the GPUs 1650 which can be compiled into at least a subset of the node devices 1500 can be integrated. Those of the task routines 1211 that for the GPUs 1650 can be compiled by the processor 1350 in appropriate GPU task routines 1671 compiled by the GPUs 1650 can be executed. However, those of the task routines can 1211 that for the GPUs 1650 so can not be compiled by the processor 1350 in corresponding CPU task routines 1571 compiled by the CPUs 1550 the node devices 1500 can be executed. The processor 1350 can then be caused to the one or more CPU task routines 1571 and / or the one or more GPU task routines 1671 for the at least one initial task to the selected one of the node devices 1500 as part of the assignment of the at least one initial task.

When executing the control routine 1540 can the CPU 1550 from each of the selected ones of the node devices 1500 the one or more CPU task routines 1571 and / or the one or more GPU task routines 1671 for the at least one initial task, and at least one of the initially distributed record partitions 1131 , When a CPU task routine 1571 is received, the CPU can 1550 be initiated, the CPU task routine 1571 execute a task with the at least one received record partition 1131 perform. When a GPU task routine 1671 is received, the CPU can 1550 be initiated, the GPU task routine 1671 to the GPU 1650 within the graphics controller 1600 the node device 1500 along with the at least one received record partition 1131 forward, with the GPU 1650 through the control routine 1640 can be initiated, the GPU task routine 1671 execute a task with the at least one received record partition 1131 perform.

Regardless of whether the CPU 1550 or a GPU 1650 from each node device 1500 a task with a corresponding record partition 1131 performs, the CPU can 1550 and / or the GPU 1650 causing a corresponding of the node status 1535 update with ads which task (s) is currently being performed and / or the extent to which various resources are currently available as a result. The CPU 1550 from each node device 1500 can by another execution of the control routine 1540 be prompted for the corresponding node status 1535 returning to the coordinator 1300 transferred to.

In a further embodiment of the control routine 1340 can the processor 1350 the coordinator 1300 the ads in the recurrently received node status 1535 the currently running tasks and / or the current levels of availability of various resources within each node device 1500 to determine the time required to complete various tasks. The processor 1350 may store indications of such durations as for each task as part of the task delay data 1535 required are. If one or more of the node devices 1500 To complete previously assigned tasks, the processor may 1350 to set such stored indications of durations to determine a predetermined period of time to assign the one or more next tasks to one or more of the node devices 1500 should be delayed.

Such delays in the assignment of the next tasks may be assigned to those of the node devices 1500 in a way that allows specific record partitions 1131 that already works within the memory 1560 and or 1660 from one or more of the node devices 1500 are stored. Furthermore, this may be the time-limited storage of record partitions 1131 within the volatile memory 1561 and or 1661 use and can exchange record partitions 1131 through the net 1999 which can consume longer durations than the lengths of the delays in assigning the next tasks. Furthermore, in embodiments where certain data set partitions 1131 originally assigned to node devices containing one or more GPUs 1650 integrating both the previously assigned tasks and the next tasks using the one or more GPUs 1650 such delays in the allocation of next task instances minimize, where the next tasks are caused to be processed in other node devices that do not have a GPU 1650 integrate so that one or more CPUs 1550 be prompted to carry out the next tasks.

With reference to 13B is the assignment of tasks to selected ones of the node devices 1500 essentially the embodiment of the assignment of tasks 13A similar, except that the compilation of the task routines 1211 the analysis routine 1210 within the selected node devices 1500 instead of within the coordinator 1300 can occur. In particular, in deriving initial assignments from at least one initial task of the analysis routine 1210 to the selected node devices 1500 the processor 1350 the coordinator 1300 the task routine (s) 1211 the at least one initial task to the selected node devices 1500 instead of CPU task routines 1571 or GPU task routines 1671 to distribute. The CPU 1550 from each of the selected nodes 1500 can the compile routine 1440 execute the distributed task routine (s) 1211 to compile the appropriate CPU task routine (s) 1571 and / or GPU Task Routine (s) 1671 within the selected node devices 1500 instead of the processor 1350 the coordinator 1300 this does.

14A . 14B and 14C together illustrate an embodiment of assigning tasks to be performed as part of the execution of an analysis routine to selected ones of the node devices 1500 of the node device grid 1005 , 14A and 14B illustrate the collection and use of data with an analysis routine 1210 , the metadata 1135 a record 1130 and / or node status 1535 associated with the node devices 1500 provided to derive the assignments. 14C illustrates the distribution of task routines 1211 . 1571 and or 1671 to the selected node devices 1500 ,

With reference to 14A as shown, the control routine 1340 that through the processor 1350 the coordinator 1300 is executed, a monitoring component 1345 to cause the processor 1350 recurring the network interface 1390 operates to the node status 1535 to receive the recurring from each of the node devices 1500 of the node device grid 1005 over the net 1999 can be transmitted. In some embodiments, each of the node devices may 1500 Recurring a node status 1535 at a regular time interval and independent of other ongoing activities as a form of "heartbeat signal" to the coordinator 1300 which displays continued functionality, and communicates information about currently available resources and / or current activities. In such embodiments, a lack of reception of the node status 1535 through the coordinator 1300 from a node device 1500 when expected, as an indication of malfunction by the node device 1500 may be taken so that the resources of the node device may be considered unavailable and any task currently assigned may be to another node device 1500 be assigned to.

As also shown, the control routine 1340 also an assignment component 1341 to cause the processor 1350 Record partitions 1131 a record 1130 along with tasks of an analysis routine 1210 that with the assigned record partitions 1131 to selected ones of the node devices 1500 of the node device grid 1005 assigns. To prepare such assignments, the processor 1350 be initiated, the network interface 1390 to operate the metadata 1135 of the record 1130 as well as the analysis routine 1210 from one or more other devices via the network 1999 for use by the assignment component 1341 retrieve. In embodiments in which the distributed processing system 1000 one or more different storage devices 1100 includes (such as the exemplary distributed processing system 1000 out 11A ), the metadata can 1135 and / or the analysis routine 1210 to the coordinator 1300 over the net 1999 from the one or more storage devices 1100 to be provided. In embodiments in which the distributed processing system 1000 however, does not include such different memory devices and the node devices 1500 a distributed storage of records 1130 (such as the exemplary distributed processing system 1000 out 11B ), the metadata can 1135 and / or the analysis routine 1210 to the coordinator 1300 over the net 1999 from one or more of the node devices 1500 to be provided.

In still other embodiments, the viewing device 1800 the coordinator 1300 the analysis routine 1210 provide. In such embodiments, the viewing device 1800 provide a user interface through which the viewing device 1800 can be controlled to send a command to the coordinator 1300 over the net 1999 to transfer the execution of the analysis routine 1210 in a distributed manner to initiate an analysis with the record 1130 perform. Part of transmitting the command to the coordinator 1300 may be the transmission of the analysis routine 1210 from the viewing device 1700 to the coordinator 1300 be.

As further shown, in embodiments where the coordinating device 1300 the task routines 1211 the analysis routine 1210 compile, the compile routine 1440 through the processor 1350 and can be an analysis component 1441 include those of the task routines 1211 to identify, accessible for conversion and compilation for embarrassing parallel execution by one or more GPUs 1650 are. For this purpose, the analysis component 1441 also to the received analysis routine 1210 access, as well as the assignment component 1341 the control routine 1340 ,

With reference to 14B As shown, the node states 1535 indicate which processing, storage and / or network access resources in each node device 1500 are integrated. As mentioned earlier, the node statuses 1535 be updated on a recurring basis to also indicate the current extent of the current availability (e.g., the current percentage of usage) of each such resource.

By way of example, the node statuses 1535 To provide displays of quantities, types, versions, and / or other internal architectural details of the processors included in each of the node devices 1500 can be integrated. Thus, the node statuses 1535 for each node device 1500 the amounts of existing CPUs 1550 and / or GPUs 1650 and specify type information for each processor, including, but not limited to, instruction set compatibility, revision level, cache size (s), amount of processing cores, and / or amount of threads that can be executed per processing core. As will be explained in more detail, the provision of such information may help to provide a heterogeneous set of node devices 1500 within the node device grid 1005 to support a variety of different processors among them. Along with information about processing resources flowing into each of the node devices 1500 can be integrated, the node status 1535 be refreshed periodically to display for the current usage levels of different processors (eg different CPUs 1550 and / or GPUs 1650 within a single node device 1500 ) to provide current usage levels of threads and / or current usage levels of other particular processing features (eg, usage levels of any specialized processing cores, logic units, statements of instruction sets, various levels of cache, etc.).

Likewise by way of example, the node statuses 1535 Ads for volatile storage capacities 1561 and or 1661 and / or non-volatile memory 1563 provide in each of the node devices 1500 can be integrated. Thus, the node statuses 1535 for each node device 1500 Specify quantities and / or various parameters of storage components that are used to store each of them 1561 . 1661 and or 1563 including, but not limited to, types of interfaces used, page and / or line sizes, access speeds and / or latencies, and / or storage technologies used. Along with information regarding the memory resources that are in each of the node devices 1500 can be integrated, the node status 1535 are updated on a recurring basis to provide indications of the current usage levels of various volatile and / or non-volatile memories.

Further, by way of example, the node statuses 1535 Display Network Access Capabilities and / or Bandwidths of the Network Interface (s) 1590 provide in each of the node devices 1500 can be integrated. Thus, the node statuses 1535 for each node device 1500 provide supported wired and / or wireless network interfaces, supported protocols, input and / or output buffer sizes, and so forth. Along with information regarding the network access resources entering each of the node devices 1500 can be integrated, the node status 1535 are updated on a recurring basis to provide indications of the current usage levels of bandwidth and / or buffers for each interface.

As also shown, the metadata 1135 Provide displays for various structural features through the data elements within the record 1130 can be organized and / or accessed. For example, the metadata 1135 View the total size, the type of data structure of the record 1130 (eg binary tree, multidimensional array, linked list, etc.), data type (s) of the data elements (eg, floating point, integer, text character, etc.), aspects of Indexing schemes used to access data elements (eg, number of dimensions, labels used, etc.), and / or other structural aspects. Also, the metadata can 1135 For example, ads for various restrictions on the way in which the record is 1130 into the record partitions 1131 include, such as any partitioning scheme, into which the data elements of the record 1130 are already organized, and / or a minimum size of the smallest atomic unit of data in which the dataset 1130 can be partitioned, which still requires independent processing of the resulting record partitions 1131 allows (for example, the size of a line or draw in a two-dimensional array, etc.). As is known to those skilled in the art, such limitations may exist with regard to the manner in which the data set 1130 can be split, an upper limit on the set of record partitions 1131 into which the record 1130 can be allocated, thereby effectively setting an upper limit on the amount of node devices 1500 can be imposed on the resulting record partitions 1131 can be distributed.

The analysis routine 1210 can work in any number of task routines 1211 each containing instructions that indicate aspects of a corresponding task that result from the execution of this task routine 1211 as part of the execution of the analysis routine 1210 to be carried out. The instructions within each of the various task routines 1211 may provide indication of required processing resources (eg, whether support for a particular instruction set extension is needed) and / or storage capacities required to support data structures instantiated during execution. The analysis routine 1210 can also provide displays for an order in which the task routines 1211 to be executed. Alternatively or additionally, the analysis routine 1210 Definitions of inputs required to perform each task and / or definitions of outputs generated by each task. Providing ads of an order in which the task routines 1211 You can display dependencies between the task routines 1211 such as displaying where an output of a task routine 1211 that is present as an input to another task routine 1211 is required. The provision of indications of required inputs and / or outputs to be generated may be part of an implementation of a many-task-computing (MTC) architecture that performs the tasks of the analysis routines 1210 are expressed independently of each other, at least by their inputs and outputs, by at least a subset of the tasks of the analysis routine 1210 for an embarrassingly parallel execution.

The manner in which definitions of inputs and / or outputs, showing the order of execution of tasks, and / or displaying dependencies, may vary depending on attributes of the programming language (s) in which the analysis routine 1210 is produced. By way of example, in some embodiments, declaration headers may be for those of the task routines 1211 that are written as callable functions in a manner that follows the conventions of a selected programming language should be sufficient to provide such definitions. However, in other embodiments, a different file or other data structure may exist either within the analysis routine 1210 or the analysis routine 1210 concomitantly to provide such definitions and / or indications. In particular, an array may be generated by way of example in which each entry indicates required inputs, outputs to be generated, and / or a position within an order of execution. Regardless of the precise manner in which definitions of inputs and / or outputs and / or dependency representations and / or order are presented in a file or other data structure, in some embodiments such definitions and / or displays may be illustrative of a directed acyclic graph (DAG) of the tasks of the analysis routine 1210 form.

When executing the assignment component 1341 can the processor 1350 be initiated, the node status 1535 , the metadata 1135 and / or parts of the analysis routine 1210 to use initial assignments of at least one initial task of the analysis routine 1210 to selected ones of the node devices 1500 and an initial distribution of at least one record partition 1131 to each of the selected ones of the node devices 1500 derive. In deriving such initial assignments, the processor may 1350 be prompted to deduce the way in which the record is made 1130 in record partitions 1131 to be split, at least in part, based on any specified constraints, within the metadata 1135 may be present, and / or on the number of node devices 1500 currently having sufficient memory resources in the node status 1535 are displayed. The processor 1350 Alternatively or additionally, the executable instructions may be initiated within one or more of the task routines 1211 of the analysis routine 1210 to identify one or more particular required processing resources (eg, floating-point math, single-instruction-multiple-data (SIMD) instruction support, etc.), and may be one of the nodes 1500 Select an initial task or tasks based on views in the node status 1535 from those of the node devices 1500 that currently have such processing resources as available should be assigned. Regardless of the exact logic used in deriving the initial assignments of tasks and / or record partitions 1131 to the selected one of the node devices 1500 used, the processor can 1350 Displaying such initial assignments as part of the assignment data 1330 to save.

In embodiments in which the coordinating device 1300 the task routines 1211 the analysis routine 1210 should be able to compile the coordinating device 1300 the compilation rules 1434 for use in executing the compile routine 1440 through the processor 1350 to save. The compilation rules 1434 can different aspects of compiling task routines 1211 the analysis routines 1210 specify the CPU task routines 1571 for execution by one or more of the CPUs 1550 and / or the GPU task routines 1671 for execution by one or more of the GPUs 1650 to create. Under what, by the compilation rules 1434 may also be aspects of transforming instructions of the task routines 1211 that were not originally created to handle the embarrassingly parallel execution capabilities of the GPUs 1650 to use (by providing a relatively large amount of threads of execution) in statements that are generated to it. Thus, the compilation rules 1434 Aspects of transforming and compiling instructions from task routines 1211 originally for execution by the CPUs 1550 were generated in instructions required for execution by the GPUs 1650 be generated.

Each of the one or more CPUs 1550 from each of the node devices 1500 can use an internal processing architecture that is well-suited for the sequential processing of task routines 1211 which includes various input / output and / or branch operations involving the execution of different sets of instructions within the task routines 1211 on the results of various provisions. Each of the CPUs 1550 can each have one or more processing cores 1555 include (with reference to 12 ), each of which can support a relatively limited degree of parallel execution of instructions on a relatively limited number of threads of execution. In contrast, the one or more GPUs 1650 in at least a subset of the node devices 1500 may be present, using an internal processing architecture that allows for embarrassingly parallel processing of the task routines 1211 are considered to be well suited, including a relatively limited set of instructions for mathematical and / or bitwise operations that can be performed independently so that there are no dependencies between multiple instances of a parallel executed task routine 1211 gives. In fact, in some embodiments, each of the GPUs may 1650 to be able to support parallel processing across hundreds, thousands or even more threads of execution.

If the instructions of a task routine 1211 can be accessed to be transformed over such a large number of threads for such a parallel execution can be that of such a task routine 1211 executed task can be executed in a much shorter time. By way of example, a task may be one of the task routines 1211 whose implementation using the CPUs 1550 the node devices 1500 can take several days, instead using the GPUs 1650 to complete only one hour per day. For tasks using the more limited instruction set of GPUs 1650 can be implemented by a single one of the GPUs 1650 within a single one of the node devices 1500 the equivalent work of many CPUs 1550 over many of the nodes 1500 in less time at a much lower cost. Such an improvement in the speed of implementation becomes even greater when more than one of the GPUs 1650 within several of the node devices 1500 operate concurrently to perform a task as an embarrassingly parallel task across the numerous threads supported by each, providing an even more highly parallel way of performing this task.

It may, however, be those of the task routines 1211 the analysis routine 1210 which include certain instructions and / or perform certain operations that follow the instructions of these task routines 1211 for such embarrassingly parallel execution by the GPUs 1650 non-convertible and non-compilable. As mentioned earlier, the GPUs 1650 support a relatively limited instruction set. By way of example, many types of input / output operations must necessarily be performed in a single thread fashion, as various protocol handshakes and / or other features impose sequential performance of steps. As a result, it is envisaged that the analysis routine 1210 both task routines 1211 where are the instructions for the conversion and compilation for the embarrassingly parallel execution by the GPUs 1650 are accessible, as well as task routines 1211 in which the instructions are not such that they are to be executed by the CPUs 1550 can be compiled.

The compilation rules 1434 can include a list of statements that, if they are in a task routine 1211 at least not the conversion and compilation of the instructions of the task routine 1211 Prevent a corresponding GPU task routine 1671 to create in which the instructions perform the task task task 1211 as a painstakingly parallel task using many threads from at least one GPU 1650 in the execution cause. Alternatively or additionally, the compilation rules 1434 include a list of instructions that, if they are within a task routine 1211 are present, such a transformation of the task routine 1211 Prevent such embarrassing parallel execution by the GPUs 1650 to support. In addition, the compilation rules 1434 specify certain circumstances under which certain instructions that would otherwise not prevent such conversion, in a task routine 1211 can be used (e.g., as part of an input / output operation or other operation) to prevent such conversion.

For those of the task routines 1211 the analysis routine 1210 that have instructions that can support conversion to other instructions that are compilable to the task (s) as embarrassing parallel task (s) on the GPUs 1650 can perform the compile rules 1434 specify different rules for performing such a transformation. By way of example, the compilation rules 1434 include one or more index value conversion rules used in loops that should be instantiated in a way that is accessible to embarrassingly parallel execution, collisions of index values, and dependencies between instances of a task that are executed in parallel , be avoided. Alternatively or additionally, one or more rules for converting instructions implemented within a loop to cause an operation to be performed multiple times sequentially may be instructions that cause multiple executions of that operation in parallel across multiple threads of at least one of the GPUs 1650 occur, give. In some such transformations; which include a loop in which the number of times the operation can be performed is large enough that not all can occur in parallel, the loop may not be eliminated, and instead may be converted to use fewer iterations if a subset of the runs are Operation occurs in parallel during each iteration. By way of example, the compilation rules 1434 One or more rules for transforming a data structure within a task routine 1211 is instantiated, whose data elements are accessed sequentially, in another data structure, on their data elements in parallel across multiple threads of at least one of the GPUs 1650 is accessed.

In some embodiments, the conversion may result in instructions of the task routine 1211 that were originally created in a particular programming language, are converted into other statements that are generated in the same programming language. However, in some such embodiments, the instructions that are defined for use in the particular language may be supplemented by one or more additional instructions associated with aspects of the internal architecture common to the GPU 1650 are specific (e.g., certain types of operands, certain types for returned values, organization of bits and / or bytes of variables for register implementation, etc.). In other embodiments, the conversion may result in instructions of the task routine 1211 that were originally created in one programming language, are converted into other instructions that are generated in a different programming language. In some such embodiments, the other programming language may be from a provider of the GPUs 1650 have been created to provide a programming language designed to match the features of the GPUs 1650 to use optimally.

The following is an example of converting parts of a task routine 1211 represented by the processor 1350 the coordinator 1300 and / or one or more CPUs 1550 the node devices 1500 can be carried out. First, parts of an example task routine 1211 produced in FCMP, a programming language offered in various products by SAS Institute Inc., Cary, NC, USA. In particular, for example, the declarations of the sub-routine "income_statement_kernel" and the data structure "d_matrix_is" have been converted between the programming languages FCMP and C.

Next, corresponding parts of a corresponding GPU task routine are shown in the conversion process executed by the processor 1350 or one or more of the CPUs 1550 is performed is generated. In particular, the above parts have become the example task routine created in FCMP 1211 To use a programming language extension that is part of the Compute Unified Device Architecture (CUDA) programming model to transform functions of GPUs (such as the GPUs 1650 ), which is distributed by Nvidia Corporation of Santa Clara, CA, a provider of GPUs called the GPUs 1650 can be used.

As those skilled in the art can readily appreciate, it is often the case that programming code originally created for execution using a first processing architecture is likely to be more efficiently executed by a first architecture processor than programming code originally intended for execution under Using a second processing architecture and then converted for execution using the first processing architecture. Nevertheless, tests have confirmed that by transforming the task routines 1211 through the compile routine 1440 significant increases in the speed of performing some tasks of an analysis routine 1210 can be realized by the generation for execution by the CPUs 1550 until creation for embarrassing parallel execution by the GPUs 1650 ,

The following is a table comparing execution times from testing an exemplary stochastic calculation. Such a calculation may be a task that is considered one of the task routines 1211 within an exemplary analysis routine 1210 is implemented. As shown, if the stochastic calculation in SAS Macro Code (another programming language offered in various products by SAS Institute Inc.) was executed by a CPU (eg, one of the CPUs 1550 from one of the nodes 1500 ), the resulting amount of time required to execute is measured at about 93,600 seconds. Instead, the same stochastic calculation is written in FCMP for execution by multiple CPUs (for example, several of the CPUs 1550 within one or more of the nodes 1500 ) resulted in a measured execution time of 763 seconds. However, if the same stochastic computation FCMP code is then scrupulously executed in parallel by one of the GPUs offered by Nvidia Corporation (eg, one of the GPUs 1650 one of the nodes 1500 ) was converted to CUDA, resulting in a measured execution time of 73 seconds. Even though 73 Seconds slower than the 11 second execution time that is achieved when the same stochastic calculation for scrupulous parallel execution by one of the GPUs offered by Nvidia Corporation is written directly to CUDA is the 73 second execution time measured by the Using the conversion from FCMP to CUDA still achieves a very significant improvement over the measured execution time of 763 seconds achieved by the execution of the FCMP code by multiple CPUs. Used technology Stochastic calculation - time in seconds CPU SAS macro code 93,600 Code written in FCMP 763 GPU Code written in FCMP, and then converted to CUDA 73 Code written directly in CUDA 11

Thus, as can be seen from these measured execution times, such use of the conversion of code to allow compilation for such embarrassingly parallel execution allows the personnel, who lack the skills or training, to have the task routines 1211 the analysis routine 1210 natively write in CUDA, but still to take advantage of an embarrassingly parallel execution of the code that they can write.

In embodiments in which the coordinating device 1300 the task routines 1211 the analysis routine 1210 should be able to compile, the processor 1350 the analysis component 1441 as part of deriving the initial allocations of record partitions 1131 and perform the initial task (s). In particular, the processor can 1350 through the analysis component 1441 be initiated every task routine 1211 the analysis routine 1210 to analyze those of the task routines 1211 identify where the instructions are accessible for conversion and compilation into appropriate GPU task routines 1671 are to do an embarrassingly parallel execution of their respective tasks by the GPUs 1650 to enable. That way, the processor can 1350 caused the ads within the compile rules 1434 which statements prevent such conversions and / or which instructions do not prevent such conversions. The processor 1350 can then be made the results of such an analysis of the task routines 1211 in the selection of node devices 1500 to be used as part of the derivation of the initial assignments. In particular, if the analysis of the task routines 1211 to a determination by the processor 1350 drives that none of the task routines 1211 in appropriate GPU task routines 1671 can be compiled, the processor 1350 the selection of node devices 1500 to restrict those to one or more of the CPUs 1550 integrate, there from the task routines 1211 no GPU task routines 1671 be generated. If the analysis of the task routines 1211 However, this leads to a determination that some of the task routines 1211 in appropriate GPU task routines 1671 While others can not be compiled, then the processor can 1350 the selection of node devices 1500 restrict to those who have both one or more CPUs 1550 as well as one or more GPUs 1650 integrate and therefore the execution of both the CPU task routines 1571 as well as the GPU task routines 1671 can support from different of the task routines 1211 be generated. Further, if the analysis of the task routines 1211 leading to a determination that all of the task routines 1211 into the appropriate GPU task routines 1671 can be compiled, the processor 1350 the selection of node devices 1500 restrict to those who have one or more of the GPUs 1650 integrate.

With reference to 14C may in embodiments in which the coordinating device 1300 the task routines 1211 the analysis routine 1210 not compiled, the processor 1350 through the assignment component 1341 the control routine 1340 be initiated, the network interface 1390 to run the task routine (s) 1211 the assigned initial tasks to the selected node devices 1500 to distribute. In such embodiments, each of the selected node devices 1500 the task routine (s) 1211 to each of the selected nodes 1500 is distributed independently to the corresponding CPU task routine (s) 1571 and / or GPU Task Routine (s) 1671 Compile, in preparation for performing the initial task (s) that each of the selected nodes 1500 is assigned / are.

Alternatively, in embodiments where the coordinating device 1300 the task routines 1211 the analysis routine 1210 should compile, so that the coordinating device 1300 the compile routine 1440 for execution by the processor 1350 stores, the compile routine 1440 a compilation component 1444 to perform the compilation of the task routines 1211 lock in. In particular, the processor can 1350 through the compilation component 1444 be initiated, at least one or more task routines 1211 Compile the one or more initial tasks associated with the selected one of the node devices 1500 have been assigned. For a task routine 1211 that from the processor 1350 (during the execution of the analysis component 1441 ) has been determined inaccessible to be converted and compiled to a corresponding GPU task routine 1671 The processor can generate 1350 be initiated, the task routine 1211 without compiling such a conversion to a corresponding CPU task routine 1571 to create. For a task routine 1211 that from the processor 1350 was determined to be accessible for conversion and compilation to a corresponding GPU task routine 1671 The processor can generate 1350 be caused to effect such conversion and compilation, the corresponding GPU task routine 1671 to create.

In some embodiments, where a task routine 1211 from the processor 1350 was determined to be accessible for conversion and compilation to a corresponding GPU task routine 1671 The processor can generate 1350 through the compilation component 1444 causing more than one compilation of the same task routine 1211 perform. For example, the processor 1350 In addition, in some such embodiments, the same task routine 1211 compile in addition to the appropriate GPU task routine 1671 also a corresponding CPU task routine 1571 to create. This may be considered desirable to address a situation where there is an insufficient number of available node devices 1500 There may be one or more of the GPUs 1650 integrate so that the task (s) of the task routine 1211 from one or more of the node devices 1500 must be performed based on the execution of the corresponding CPU task routine 1571 through one or more CPUs 1550 , Alternatively or additionally, this may be considered desirable to address a situation in which a node device 1500 containing one or more GPUs 1650 integrated when running the appropriate GPU task routine 1650 fails and performing the task (s) of the task routine 1211 with the respective record partition (s) 1131 , this node device 1500 is assigned to another node device 1500 must be assigned, no GPU 1650 integrated.

As another example of more than one compilation of the same task routine 1211 can the node device grid 1005 in some of these embodiments, a heterogeneous set of node devices 1500 Include different GPUs 1650 integrate that does not share an instruction set, allowing the generation of a single corresponding GPU task routine 1671 for all different GPUs 1650 can not be possible. Thus, the processor can 1350 cause several conversions and compilations of the same task routine 1211 in each of the different versions of the GPU task routine 1671 perform for each of the various GPUs 1650 that under the node devices 1500 of the node device grid 1005 required are.

However, in some of the embodiments where the coordinating device 1300 the task routines 1211 the analysis routine 1210 compile, no matter if the analysis of the task routines 1211 to identify those who are compiling for one or more different GPUs 1650 are accessible, take place and / or whether such a compilation for generating corresponding GPU task routines 1671 be conditioned, whether there are indications that any node devices 1500 present are a GPU 1650 integrate. In particular, in a situation where the current node status 1535 indicate that none of the node devices containing one or more of the GPUs 1650 integrate, are currently available to ever assigned a task, the processor 1350 through the assignment component 1341 be induced by performing an analysis of the task routines 1211 to determine if any for compilation to be executed by a GPU 1650 are accessible. Alternatively or additionally, in embodiments in which various node devices 1500 different types of GPUs 1650 integrate, allowing multiple compilations to generate GPU task routines 1671 for all of the different types of GPUs 1650 are required to encounter a situation where the node status 1535 show all of the GPUs 1650 one of the types currently not available for performing tasks. In such a situation, the processor can 1350 be caused by the assignment routine, by the compilation of any task routine 1211 foresee a GPU task routine 1671 for execution by GPUs 1650 of this particular type.

The generation of the GPU task routines 1671 in a way that the conversion of the instructions of the task routines 1211 includes in other instructions the embarrassingly parallel processing features of the GPUs 1650 use, followed by compilation, may be considered desirable to improve the ease with the embarrassingly parallel features of the GPUs 1650 can be used. Through such use of instructions conversions staff can be given the opportunity to use the GPUs 1650 to be used in at least some of the node devices 1500 can be present without the need to follow the instructions of the task routines 1211 to architect and write in a manner designed for such embarrassingly parallel execution. In other words, this staff may be spared the need to acquire the additional skills and experience to architect the instructions and write the tasks of the analysis routine 1210 implement in a way that is embarrassing for parallel execution by the GPUs 1650 is designed. Additionally, if the node device grid 1005 node devices 1500 includes, the different of the GPUs 1650 Further, such personnel will be spared the need to architect and write different versions of instructions that will perform the tasks of the analysis routine 1210 implement in a manner appropriate to the different peculiarities of the way in which a highly parallel execution by each of the different ones of the GPUs 1650 is provided is designed.

15A and 15B together illustrate one embodiment of performing tasks, the selected ones of the node devices 1500 of the node device grid 1005 are assigned. 15A illustrates aspects of preparing for the implementation that involves compiling a task routine 1211 within an exemplary node device 1500 can include, and 15B illustrates aspects of performing a task within the example node device 1500 by executing the corresponding CPU task routine 1571 or equivalent GPU task routine 1671 ,

With reference to 15A may, as previously discussed, in embodiments in which the distributed processing system 1000 one or more different storage devices 1100 includes (such as the exemplary distributed processing system 1000 out 11A ), one or more record partitions 1131 to the node devices 1500 over the net 1999 be provided, including to the illustrated exemplary node device 1500 , However, in embodiments where the distributed processing system 1000 does not include such different memory devices and the node devices 1500 a distributed storage of the records 1130 (such as the exemplary distributed processing system 1000 out 11B ), one or more record partitions 1131 with which a task within the exemplary node device 1500 is to be performed already within the exemplary node device 1500 be saved. Otherwise, one or more record partitions 1131 to the exemplary node device 1500 over the net 1999 from another of the node devices 1500 to be provided.

As previously discussed, in embodiments in which the coordinating device 1300 the task routines 1211 the analysis routine 1210 not compiled, the coordinator 1300 the one or more task routines 1211 the one or more assigned initial tasks to the selected node devices 1500 to distribute. In such embodiments, and as illustrated, each of the node devices may 1500 the compile routine 1440 for execution by a CPU 1550 from each of the node devices 1500 to save. Thus, after receiving a task routine 1211 from the coordinator 1300 a CPU 1550 the exemplary node device 1500 the compile routine 1440 Run to the task routine first 1211 to analyze whether it is accessible for conversion and compilation to a corresponding GPU task routine 1671 to create. If so, the CPU can 1550 such a conversion and compilation of the task routine 1211 to perform the appropriate GPU task routine 1671 for execution by a GPU 1650 the exemplary node device 1500 to create. If the task routine 1211 However, not as accessible, the CPU can 1550 be initiated, the task routine 1211 without compiling such conversion to the appropriate CPU task routine 1571 to create.

Alternatively, in embodiments where the coordinating device 1300 the task routines 1211 the analysis routine 1210 compiled, the coordinator 1300 the one or more corresponding CPU task routines 1571 and / or GPU task routines 1671 distribute that within the coordinator 1300 from the one or more task routines 1211 were generated. As a result, the example node may 1500 a CPU task routine 1571 that from the CPU 1550 to be executed, or a GPU task routine 1671 that from the GPU 1650 to be executed, instead of the corresponding task routine 1211 one of which is generated.

As shown, the control routine 1640 by the GPU 1650 is executed, an execution component 1641 include to the GPU 1650 to initiate the appropriate GPU task routine 1671 if the appropriate GPU task routine 1671 from the task routine 1211 the initial task, that of the exemplary node device 1500 is assigned (whether within the coordinator 1300 or the exemplary node device 1500 ). As also shown, the control routine 1540 that through the CPU 1550 is executed, but an execution component 1541 include to the CPU 1550 to initiate the appropriate CPU task routine 1571 if the appropriate GPU task routine 1671 not from the task routine 1211 the initial task, that of the exemplary node device 1500 is assigned, so that the generation of the corresponding CPU task routine 1571 necessary is.

With reference to 15B , and with brief reference to 12 where the GPU 1650 through the execution component 1641 causes the GPU task routine 1671 to execute the assigned initial task with a record partition 1131 can perform pages both from the GPU task routine 1671 as well as the record partition 1131 between the non-volatile memory 1563 and the volatile memory 1661 with which the GPU 1650 is coupled, exchanged. Alternatively, where the CPU 1550 through the execution component 1541 is the CPU task routine 1571 to execute the assigned initial task with a record partition 1131 can perform pages from both the CPU task routine 1571 as well as the record partition 1131 between the non-volatile memory 1563 and the volatile memory 1561 with which the CPU 1550 is coupled, exchanged. As previously related to 12 Everyone can discuss the volatile memory 1561 and 1661 a much faster access than the non-volatile memory 1563 allow data and / or routines stored therein so that the CPU 1550 and the GPU 1650 can execute more efficient routines and perform operations on data, each within the volatile memory 1561 and 1661 are stored. However, everyone can use the volatile memory 1561 and 1661 a much smaller storage capacity than the non-volatile memory 1563 exhibit. As a result, the situation can occur repeatedly if significantly more (if not all) of a relatively large data set partition 1131 within the non-volatile memory 1563 is storable while only a relatively small part of this record partition 1131 within one of the volatile memory 1561 and 1661 is storable, whereby the use of a page exchange is required.

As shown, the control routine 1540 also a status component 1543 include to the CPU 1550 to initiate the network interface 1590 of the exemplary node 1500 to periodically update the current status of the processing, storage, and / or network access resources of the example node 1500 as updated instances of node status 1535 to the coordinator 1300 transferred to. As previously related to 12 discussed, the GPU can 1650 within the graphics controller 1600 no access to the network interface 1590 have and / or can not run the required driver routines to the network interface 1590 to operate directly. Thus, the CPU can 1550 through the status component 1543 be prompted to periodically display the status the GPU 1650 and / or volatile memory 1661 from the GPU 1650 to add them to the status indicators in the transmitted node status 1535 are included.

16 FIG. 12 illustrates an embodiment of an assignment of next tasks to be performed as part of the execution of an analysis routine to selected ones of the node devices 1500 of the node device grid 1005 , As previously in connection with 14A -Federation 15B discussed, the processor can 1350 the coordinator 1300 through the monitoring component 1545 be initiated, the network interface 1390 operate recurrently to the node status 1535 to receive the recurring from the node devices 1500 of the node device grid 1005 over the net 1999 be transmitted.

In some embodiments, the processor 1350 also by the monitoring component 1545 be prompted to display the time periods required for completion of tasks involving the node devices 1500 are assigned, required, maintained and updated on a recurring basis. In some such embodiments, the task delay data may be 1335 include a separate per-task time for completion, recurring as an average of the time durations for each of the multiple node devices 1500 are required to complete the same task is calculated. In embodiments in which the task may be different to the node devices 1500 in which they are used with different processor types (eg a CPU 1550 vs. a GPU 1650 or different types of GPUs 1650 ) can be executed within the task delay data 1335 for each different type of processor used to perform a task, different average times are maintained. Alternatively or additionally, as a mechanism to account for differences in resources (including processing resources) between the node devices 1500 Averages of relative differences in durations for each node device 1500 to complete assigned tasks compared to other node devices 1500 in the task delay data 1335 be included.

In such embodiments, the processor 1350 through the assignment component 1541 causes the stored indications of time durations to be completed for assigned tasks within the task delay data 1335 to use one or more time periods to deduce the processor 1350 the assignment of a next task with a specific record partition 1131 to a node device 1500 may not delay the particular record partition 1131 stored in it. For multiple time periods of delay, each of them may be based on a different average of completion times, within the task delay data 1335 stored for a different task. When differences in resources between the node devices 1500 may be accounted for by averages of relative differences in durations, such averages may be from the processor 1350 used to time each of the delay periods for a particular node device 1500 to modify.

17A . 17B and 17C taken together in the order of 17A to 17C 12 illustrate an embodiment of performing tasks among a plurality of selected ones of the node devices 1500 of the node device grid 1005 , In particular, ask 17A C is an exemplary triad of node devices 1500x . 1500y and 1500Z This is a first task with an example record 1130a and then perform a second task in conjunction with the first task with the same example record 1130a carry out. In this way, each of the exemplary node devices should 1500x -Z another example record 1130b and then another example record 1130c produce. It should be noted that this example is based on the only three node devices 1500x -Z based and the three records 1130a C includes, which may be small enough in size to access the only three node devices 1500 being a deliberately simplified example and discussed herein for purposes of explanation and understanding, and should not be construed as limiting. In particular, it is envisioned that embodiments are likely to perform tasks with significantly larger data sets 1130 and thus probably the use of significantly more node devices 1500 include.

Starting from 17A became each of the three node devices 1500x -Z assigned the first task with a corresponding one of three record partitions 1131ax -Az of the record 1130a to perform a respective one of the three record partitions 1131bx -Bz of the record 1130b to create. In addition, each of the three node devices 1500x -Z with its corresponding one of the three record partitions 1131ax -Az of the record 1130a from the one or more storage devices 1100 over the net 1999 Mistake.

As illustrated, one or more processors of the node device are running 1500x (eg one or more CPUs 1550 and / or one or more GPUs 1650 ) nor the execution of the first task with the record partition 1131ax so that the record partition 1131bx still within the node device 1500x is produced. Additionally, it may be that performing the first task with the record partition 1131ax sufficient processing and / or storage resources of the node device 1500x consumes that it is considered that the node device 1500x may have insufficient processing and / or memory resources to be assigned to perform another task until the first task with the record partition 1131ax was completed. When performing the first task with the record partition 1131ax to the record partition 1131bx may generate at least a subset of each of these partitions within the non-volatile memory 1563 the node device 1500x get saved. In addition, pages of these two partitions can be used between the non-volatile memory 1563 and one or the other of the volatile memory 1561 or 1661 be exchanged, depending on whether the first task by one or more CPUs 1550 or one or more GPUs 1650 the node device 1500x is carried out.

As also shown, one or more processors have the node device 1500y already performing the first task with the record partition 1131ay completed, leaving the record partition 1131by already within the node device 1500y and then stored for storage by the one or more memory devices 1100 was transferred. In addition, after completing this execution of the first task, the node device became 1500y assigned the second task with the record partition 1131ay to perform a record partition 1131cy of the record 1130c to create. As illustrated, one or more processors of the node device are running 1500y still performing the second task with the record partition 1131ay so that the record partition 1131cy still within the node device 1500y is produced. Additionally, it may be that performing the second task with the record partition 1131ay sufficient processing and / or storage resources of the node device 1500y consumes that it is considered that the node device 1500y may have insufficient processing and / or memory resources to be assigned to perform another task until the second task with the record partition 1131ay was completed. As the first task with the record partition 1131ay was performed to the record partition 1131by can generate at least a subset of the partition 1131by for a limited period of time within the non-volatile memory 1563 the node device 1500y stay saved. When performing the second task with the record partition 1131ay to the record partition 1131cy may generate at least a subset of each of these partitions within the non-volatile memory 1563 the node device 1500x get saved. Additionally, pages of these two record partitions can be between the non-volatile memory 1563 and one or the other of the volatile memory 1561 or 1661 be exchanged, depending on whether the second task by one or more CPUs 1550 or one or more GPUs 1650 the node device 1500y is carried out. Furthermore, as a result of such exchanges, it may be unlikely that any page of the record partition 1131by still within the volatile memory 1561 or 1661 is stored.

As further illustrated, one or more processors of the node device 1500Z already performing the first task with the record partition 1131az completed, leaving the record partition 1131bz already within the node device 1500Z and then stored for storage by the one or more memory devices 1100 was transferred. Similarly, the one or more processors of the node device 1500Z already performing the second task with the record partition 1131az completed, leaving the record partition 1131cz also already within the node device 1500Z and then stored for storage by the one or more memory devices 1100 was transferred. Therefore, in contrast to the node devices 1500x and 1500y be considered that the node device 1500Z has enough available processing and memory resources to allow the node device 1500Z can be assigned to perform another task. As a result of performing the first task with the record partition 1131az to generate the record partition 1131bz and as a result of performing the second task with the record partition 1131az to generate the record partition 1131cz may be at least a subset of one or more of the record partitions 1131az . 1131bz and 1131cz for a limited period of time within the non-volatile memory 1563 the node device 1500Z stay saved. Also, as a result of performing the second task more recently than the first task, pages of one or both of the record partitions 1131az and 1131cz still within one or the other of the volatile memory 1561 or 1661 be stored, depending on whether the second task by one or more CPUs 1550 or one or more GPUs 1650 the node device 1500Z was carried out. It may, however, be unlikely that any side of the partition 1131bz still inside the volatile memory 1561 or 1661 is stored.

Thus, the node device 1500Z , under the circumstances just described and as indicated by dashed lines in FIG 17B presented, available so that it could be assigned to the second task with the record partition 1131ax of the record 1130a to perform the record partition 1131cx of the record 1130c to create. However, as already discussed above, it is the node device 1500x originally with the record partition 1131ax from the one or more storage devices 1100 was provided. Therefore, the record partition becomes 1131ax already inside the non-volatile memory 1563 the node device 1500x saved, leaving the record partition 1131ax not again over the network 1999 from the one or more storage devices 1100 would have to be transmitted if the node device 1500x could be assigned to the second task with the record partition 1131ax perform. Also, as already mentioned, due to the ongoing execution of the first task with the record partition 1131ax within the node device 1500x currently exchanging pages of the record partition 1131ax between the non-volatile memory 1563 and either the volatile memory 1561 or 1661 , Thus, one or more pages of the record partition 1131ax currently within the volatile memory 1561 or 1661 the node device 1500x stored, and the relatively fast assignment of the implementation of the second task with the record partition 1131ax to the node device 1500x after the node device 1500x completing their execution of the first task may be the temporary storage of that one or more pages within the volatile memory 1561 or 1661 use, which makes the node device 1500x can be made to start the implementation of the second task much faster.

Therefore, as indicated by dashed lines in 17B shown, despite the availability of the node device 1500Z to be assigned to the second task with the record partition 1131ax to perform the node device 1500Z not assigned to do this. Instead, the processor becomes 1350 the coordinator 1300 through the assignment component 1341 caused (with reference to 16 ), the assignment of performing the second task with the record partition 1131ax to any of the node devices 1500x -Z for a period of time to delay the node device 1500x to provide a way of carrying out the first task with the record partition 1131ax complete.

As discussed earlier, in some embodiments, the duration of the time period of such a delay may be indicative of indications of the recurrently received node status 1535 based on how long it took for one or more nodes to complete the same task for which the assignment is being delayed (eg, based on an average generated from the times required for one or more nodes) to complete the same task). However, as has also been discussed, the time period of the delay may also be based on determinations of differences in the time periods that the node needs, given the opportunity to complete a task, compared to the time periods that other nodes require to complete the task complete same task. As mentioned earlier, consideration of such differences in durations may be considered desirable to account for differences that exist in the various resources that exist in different node devices 1500 are integrated. Thus, in the example 17A -C the time period of the delay to the node device 1500x a way to give the first task with the record partition 1131ax to complete, on the time duration, for one or both of the node devices 1500y and 1500Z was required to complete the first task with their respective record partitions 1131ay and 1131az (eg an average of it), and / or on differences in durations that the node device 1500x needed to complete tasks, compared to the time periods that the node devices 1500y and or 1500Z need to complete the same tasks are based.

Regardless of the exact manner in which the time period of the delay in allocating the performance of the second task with the record partition 1131ax to one of the node devices 1550x -Z is derived, the processor can 1350 the coordinator 1300 a bar 1351 Use to monitor the timing in the processor 1350 can be integrated (with reference to 16 ) to determine when the time period of the delay has completely expired. If the time period of the delay has passed completely before the node device 1500x their performing the first task with the record partition 1131ax then the processor can 1350 through the assignment component 1341 be prompted to perform the second task with the record partition 1131ax to the node device 1500Z assign. If, as in 17C shown, the node device 1500x however, its performing the first task with the record partition 1131ax closes, before the time period of the delay expires, then the processor can 1350 be prompted to perform the second task with the record partition 1131ax to the node device 1500x assign. As already mentioned, such an assignment of the second task with the record partition 1131ax to the node device 1500x at least the storage of at least a part of the record partition 1131ax within the non-volatile memory 1563 if not also saving one or more pages of the record partition 1131ax within the volatile memory 1561 or 1661 ,

18A . 18B and 18C taken together in the order 18A to 18C 15 illustrate another embodiment of performing tasks among a plurality of selected ones of the node devices 1500 of the node device grid 1005 , In particular, ask 18A -C the same exemplary triad of node devices 1500x . 1500y and 1500Z This is a first task with an example record 1130a perform an exemplary record 1130b and then a second task with the example record 1130b performed by performing the first task. In performing the second task with the example record 1130b Let each of the exemplary node devices 1500x -Z yet another exemplary record 1130c produce. As in 17A -C, it should be noted that this example is based on only three node devices 1500 based and the three records 1130a C includes, which may be small enough in size to access the only three node devices 1500x Being a deliberately simplified example and discussed herein for purposes of explanation and understanding, and should not be considered as limiting. As mentioned earlier, it is envisioned that embodiments are likely to perform tasks with significantly larger datasets 1130 and thus probably the use of much more of the node devices 1500 include.

Starting from 18A became each of the three node devices 1500x -Z assigned the first task with a corresponding one of three record partitions 1131ax -Az of the record 1130a to perform a respective one of the three record partitions 1131bx -Bz of the record 1130b to create. As illustrated, all three of the node devices 1500x -Z completed its completions of the first task, and each has its corresponding the record partitions 1131bx -Bz to the one or more storage devices 1100 for storage as a data record 1130b transfer. To enable these implementations of the first task, each of the three node devices became one 1500x -Z with its corresponding one of the three record partitions 1131ax -Az of the record 1130a from the one or more storage devices 1100 over the net 1999 Mistake.

After their first task completes, each of the knot devices became 1500y and 1500Z assigned the second task with the record partitions 1131by and 1131bz perform the node devices 1500y and 1500Z each generated as a result of their implementations of the first task. After completion of the first task became the node device 1500x however, assigned to perform a task of another and unrelated parser. As previously discussed, the assignment of tasks of different and unrelated analysis routines may be done in embodiments in which the node device grid 1005 is shared to the extent that several unrelated analysis routines are using the node devices 1500 be carried out simultaneously.

As illustrated, one or more processors of the node device are running 1500x (eg one or more CPUs 1550 and / or one or more GPUs 1650 ) nor the execution of the task of the unrelated analysis routine. In addition, performing the task of the unrelated analysis routine may be sufficient processing and / or memory resources of the node device 1500x consumes that it is considered that the node device 1500x may have insufficient processing and / or memory resources to be assigned to perform another task until the unrelated parsing routine task is completed. As the first task with the record partition 1131ax was performed to the record partition 1131bx can generate at least a subset of one or both of the partitions 1131ax and 1131bx for a limited period of time within the non-volatile memory 1563 the node device 1500x stay saved. In performing the task of the unrelated parsing routine, an exchange of pages of an unrelated record partition between the nonvolatile memory 1563 and one or the other of the volatile memory 1561 or 1661 Depending on whether the task of the unrelated analysis routine is performed by one or more CPUs 1550 or one or more GPUs 1650 the node device 1500x was carried out. Furthermore, as a result of such exchanges, it may be unlikely that any page of the record partitions 1131ax or 1131bx still within the volatile memory 1561 or 1661 is stored.

As illustrated, one or more processors of the node device are running 1500y still performing the second task with the record partition 1131by so that the record partition 1131cy still within the node device 1500y is produced. Additionally, it may be that performing the second task with the record partition 1131by sufficient processing and / or storage resources of the node device 1500y consumes that it is considered that the node device 1500y may have insufficient processing and / or memory resources to be assigned to perform another task until the second task with the record partition 1131ay was completed. As the first task with the record partition 1131ay was performed to the record partition 1131by can generate at least a subset of the partition 1131ay for a limited period of time within the non-volatile memory 1563 the node device 1500y stay saved. When performing the second task with the record partition 1131by to the record partition 1131cy may generate at least a subset of each of these partitions within the non-volatile memory 1563 the node device 1500x get saved. Additionally, pages of these two record partitions can be between the non-volatile memory 1563 and one or the other of the volatile memory 1561 or 1661 be exchanged, depending on whether the second task by one or more CPUs 1550 or one or more GPUs 1650 the node device 1500y is carried out. Furthermore, as a result of such exchanges, it may be unlikely that any page of the record partition 1131ay still within the volatile memory 1561 or 1661 is stored.

As further illustrated, one or more processors of the node device 1500Z already performing the second task with the record partition 1131bz completed, leaving the record partition 1131cz already within the node device 1500Z and then stored for storage by the one or more memory devices 1100 was transferred. Therefore, in contrast to the node devices 1500x and 1500y be considered that the node device 1500Z has enough available processing and memory resources to allow the node device 1500Z can be assigned to perform another task. As a result of performing the first task with the record partition 1131az to generate the record partition 1131bz and as a result of performing the second task with the record partition 1131bz to generate the record partition 1131cz may be at least a subset of one or more of the record partitions 1131az . 1131bz and 1131cz for a limited period of time within the non-volatile memory 1563 the node device 1500Z stay saved. Also, as a result of performing the second task more recently than the first task, pages of one or both of the record partitions 1131bz and 1131cz still within one or the other of the volatile memory 1561 or 1661 be stored, depending on whether the second task by one or more CPUs 1550 or one or more GPUs 1650 the node device 1500Z was carried out. However, it can be unlikely that any side of the partition 1131az still inside the volatile memory 1561 or 1661 is stored.

Thus, under the circumstances just described and as indicated by dashed lines in FIG 18B shown, the node device 1500Z available, allowing them to perform the second task with the record partition 1131bx of the record 1130b could be assigned to the record partition 1131cx of the record 1130c to create. However, as already discussed above, it is the node device 1500x that was originally the record partition 1131bx produced. Therefore, the record partition becomes 1131bx already inside the non-volatile memory 1563 the node device 1500x saved, leaving the record partition 1131bx not over the net 1999 from the one or more storage devices 1100 (or from the node device 1500x ) would have to be transmitted if the node device 1500x could be assigned to the second task with the record partition 1131bx perform. Thus, the relatively fast allocation of the performance of the second task with the record partition 1131bx to the node device 1500x after the node device 1500x their completion of the task of the unrelated analysis routine completes the temporary storage of the record partition 1131bx within the non-volatile memory 1563 the node device 1500x use what it's the node device 1500x make it possible to start the second task much faster.

Therefore, as indicated by dashed lines in 18B shown, despite the availability of the node device 1500Z to be assigned to the second task with the record partition 1131bx to perform the node device 1500Z not assigned to do this. Instead, the processor becomes 1350 the coordinator 1300 through the assignment component 1341 caused (with reference to 16 ), the assignment of performing the second task with the record partition 1131bx to any of the node devices 1500x -Z for a period of time to delay the node device 1500x to provide a way of performing their task of unrelated Complete analysis routine. Similar to the embodiment of 17A -C, if the time period of the delay has passed completely before the node device 1500x is able to complete their execution of the task of unrelated analysis routine, then the processor can 1350 through the assignment component 1341 be initiated (with reference to 16 ), performing the second task with the record partition 1131bx to the node device 1500Z assign. If, as in 18C shown, the node device 1500x however, completing its execution of the task of unrelated parsing routine before the time period of the delay expires, then the processor may 1350 be prompted to perform the second task with the record partition 1131bx to the node device 1500x assign.

19 illustrates an embodiment of a logic flow 2100 , The logic flow 2100 may be representative of some or all of the operations performed by one or more of the embodiments described herein. In particular, the logic flow 2100 Illustrate operations by the processor 1350 during the execution of the control routine 1340 and / or the compile routine 1440 and / or other component (s) of the coordinator 1300 be performed.

at 2110 For example, a processor may be a coordinating device of a distributed processing system (eg, the processor 1350 the coordinator 1300 of the distributed processing system 1000 ) Analyze node status that the coordinator recursively receives from node devices of a grid of node devices (eg, node statuses 1535 the node devices 1500 of the node device grid 1005 ) to determine which processing resources are available within each of the node devices. As discussed earlier, the node states 1535 returning through the knot devices 1500 to the coordinator 1300 indicate which processing resources are in each of the node devices 1500 are integrated (eg which CPU (s) 1500 and / or GPU (s) 1600 including, what processing features are provided by each, etc.) and / or to what extent these processing resources are currently within each of the node devices 1500 Are available.

at 2120 For example, the processor of the coordinator may check whether any GPUs are displayed as sufficiently available within any of the node devices in the node status so that a task could be assigned to these node devices to be performed by such available GPUs. If no GPUs are so available, the processor at 2122 compile all of the task routines of an analysis routine to be executed by the CPUs of the node devices (eg, the task routines 1211 an analysis routine 1210 ). at 2124 For example, the processor may prioritize the assignment of all of the tasks of the analysis routine to node devices that have been selected because they integrate only one or more CPUs and / or integrate a combination of one or more CPUs and one or more GPUs.

However, if GPUs are present, the at 2120 can be considered sufficiently available, then the processor can 2130 Analyze each of the analysis routine's task routines to determine if any of the task routines are accessible for conversion of their instructions and compilation to be performed by the GPUs of the node devices of the node device grid. If at 2140 If there are no such accessible task routines, then the processor can 2122 compile all of the task routines of the analysis routines to be executed by the CPUs of the node devices. Then the processor at 2124 again prioritizing the assignment of all of the tasks of the analysis routines to node devices that have been selected because they integrate only one or more CPUs and / or integrate a combination of one or more CPUs and one or more GPUs. As previously discussed, if none of the task routines 1211 an analysis routine 1210 for compilation for execution by a GPU 1650 the prioritization of the assignment of the tasks of such an analysis routine 1210 to node devices 1500 that only CPU (s) 1550 integrate, if possible, and / or node devices 1500 that is a combination of the CPU (s)) 1550 and the GPU (s) 1650 integrate, if necessary, deemed desirable to the assignments of tasks to node devices 1500 , the GPUs 1650 integrate, minimize, if none of the tasks of a GPU 1650 to be executed. This can help make the GPUs 1650 the node devices 1500 of the node device grid 1005 remain more readily available for use in performing tasks of another analysis routine that can be performed by GPUs.

If at 2140 however, there is at least one task routine of the analysis routine that is accessible for such conversion and compilation, at 2150 It checks whether all of the task routines of the analysis routine are so accessible. If so, the processor may be at 2152 all of the Compile task routines of the analysis routines to be executed by the GPUs of the node devices. at 2154 For example, the processor may prioritize the assignment of all of the tasks of the analysis routine to node devices that have been selected because they integrate only one or more GPUs and / or integrate a combination of one or more CPUs and one or more GPUs.

If at 2150 However, if there is a mix of task routines that are accessible and inaccessible for such conversion and compilation, then the processor may be unable to do so 2160 compile all of the analysis routine task routines that are accessible to be executed by the GPUs of the node devices. at 2162 For example, the processor may compile all of the task routines of the analysis routines that are not as accessible to be executed by the CPUs of the node devices. at 2164 For example, the processor may prioritize the assignment of all of the tasks of the analysis routine to node devices that have been selected because they integrate a combination of one or more CPUs and one or more GPUs.

20 illustrates an embodiment of a logic flow 2200 , The logic flow 2200 may be representative of some or all of the operations performed by one or more of the embodiments described herein. In particular, the logic flow 2200 Illustrate operations by the one or more CPUs 1550 during the execution of the control routine 1540 and / or the compile routine 1440 and / or other component (s) of the node devices 1500 be performed.

at 2210 For example, a CPU of a node device of a distributed processing system (eg, a CPU 1550 one of the node devices 1500 of the distributed processing system 1000 ) a task routine (eg, a task routine 1211 an analysis routine 1210 ) to determine whether it is amenable to conversion of its instructions and compilation to be performed by the one or more GPUs of the node device (e.g., one or more of the GPUs 1650 ) to be executed. As previously discussed, in embodiments where the coordination device 1300 the task routines 1211 not compiled, one or more CPUs 1550 from each of the node devices 1500 to do this. Also, such a compilation may be an analysis of each task routine 1211 that of each of the node devices 1500 is received, the one or more of the GPUs 1650 integrated to determine if compilation is to be executed by one or more CPUs 1550 or one or more GPUs 1650 to be executed.

If the task routine at 2220 not so accessible, then the CPU can 2222 compile the task routine of the parsing routine to be executed by the one or more CPUs of the node device. If the task routine at 2220 However, as accessible, then the CPU can 2230 compile the task routine of the parsing routine to be executed by the one or more GPUs of the node device.

21 illustrates an embodiment of a logic flow 2300 , The logic flow 2300 may be representative of some or all of the operations performed by one or more of the embodiments described herein. In particular, the logic flow 2300 Illustrate operations by the processor 1350 during the execution of the control routine 1340 and / or other component (s) of the coordinator 1300 be performed.

at 2310 For example, a processor may be a coordinating device of a distributed processing system (eg, the processor 1350 the coordinator 1300 of the distributed processing system 1000 ) Receive metadata of a record descriptive of a record of one or more storage devices (e.g., the metadata 1135 of one or more storage devices 1100 ). at 2312 For example, the processor may receive an analysis routine that includes multiple task routines from the one or more storage devices and / or a viewer (eg, the analysis routine 1210 that have multiple task routines 1211 from the one or more storage devices 1100 or from the viewing device 1800 including). at 2314 For example, the processor may receive the most recent transmissions of node status data from node devices of a lattice from node devices of the distributed processing system (e.g., the node statuses 1535 the node devices 1500 of the node device grid 1005 ). As previously discussed, the coordinating device 1300 in embodiments in which the node devices 1500 also as storage devices of at least one or more data sets 1130 serve, the metadata 1135 and / or the analysis routine of one of the node devices 1500 receive.

at 2320 For example, the processor may make initial allocations of record partitions (eg, record partitions 1131 ) and initial tasks, including a selection of several of them Node devices to which data set partitions are to be distributed and initial assignments of the one or more initial tasks are to be made. at 2322 For example, the processor may transmit displays via the assigned distribution of the record partitions to one or more storage devices and / or to the selected node devices. As previously discussed, the distribution of record partitions to the selected node devices may be coordinated by the one or more memory devices among the one or more memory devices, the selected node devices, and / or the coordinator device in any of a variety of ways Is used to initiate the transfer of the record partitions to the selected node devices.

at 2330 For example, the processor may transmit indications about the assignment of a first task to the selected node devices to be performed on the selected node devices with corresponding ones of the record partitions. As previously discussed, the transmission of task routines may be to the parsing routine or the transfer of CPU task routines and / or GPU task routines (eg, the task routines 1211 , the CPU task routines 1571 and / or the GPU task routines 1671 ) to the selected node devices as part of the signaling of assignments of tasks to the selected node devices.

at 2340 For example, the processor may receive, from a first node device of the selected node devices, an indication of completion of the first task with the first record partition by the first node device. As previously discussed, such displays may be considered part of the recurring transmissions of node status 1535 be transmitted to the coordinating device.

at 2350 For example, the processor may delay the assignment of a second task using a second record partition to any of the selected node devices, including the first node device, for a predetermined period of delay to a second node device of the selected node devices in which the second record partition already stored, to give a chance to become available. As previously discussed, such imposition of a delay in assigning a next task to a particular record partition may allow a node device that already stores that particular record partition to be made available to store that particular record partition within that node device by assigning the next task with the particular record partition to that node device. As already mentioned, in this way the network bandwidth resources and the time required for the transmission of the particular data record partition to another node device can be spared, thus waiting for this node device to become available.

at 2360 For example, the processor may check to see if the second node device in which the second data set partition has already been stored became available at the end of the time period of the delay. If the second node device became available within the time period of the delay, then the processor may provide an indication of the assignment to perform the second task with the second data set partition 2362 transmitted to the second node device. However, if the second node device did not become available within the time period of the delay, then the processor may provide an indication of the allocation to perform the second task with the second data set partition 2370 transmitted to the first node device.

22 illustrates an embodiment of a logic flow 2400 , The logic flow 2300 may be representative of some or all of the operations performed by one or more of the embodiments described herein. In particular, the logic flow 2400 Illustrate operations by the processor 1350 during the execution of the control routine 1340 and / or other component (s) of the coordinator 1400 be performed.

at 2410 For example, a processor may be a coordinating device of a distributed processing system (eg, the processor 1350 the coordinator 1300 of the distributed processing system 1000 ) an indication from a node device of a node device grid of the distributed processing system (e.g., one of the node devices 1500 of the node device grid 1005 ) that this node device has completed its execution of a task routine containing a record partition of a record (eg, one of the record partitions 1131 of the record 1130 ) includes. at 2412 For example, the processor may determine the amount of time that has elapsed between its transmission of a signal to the node device to perform the task with the record partition and receipt of the indication from the node device upon the completion of this performance of that task.

at 2420 For example, the processor may compare the new elapsed time with previous elapsed times needed by other node devices to perform the same task with their corresponding record partitions of the same record. at 2422 For example, the processor may check whether the new elapsed time, similar to the previous elapsed times, for these implementations of the same task are within a predetermined threshold of similarity. If not, then the processor may use the new elapsed time along with the previous elapsed times to again derive a time period of delay by which a new task assignment should be delayed in order to complete completion of the task by another node enable. As previously discussed, such determinations of a time period of delay may average a time duration of different ones of the node devices 1500 needed to perform the same task.

23 illustrates an embodiment of a logic flow 2500 , The logic flow 2500 may be representative of some or all of the operations performed by one or more of the embodiments described herein. In particular, the logic flow 2500 Illustrate operations by the processor 1350 during the execution of the control routine 1340 and / or other component (s) of the coordinator 1300 be performed.

at 2510 For example, a processor may be a coordinating device of a distributed processing system (eg, the processor 1350 the coordinator 1300 of the distributed processing system 1000 ) from a first node device of a node device grid of the distributed processing system (e.g., one of the node devices 1500 of the node device grid 1005 ) an indication of the completion of a first task with a first record partition of a record (eg, one of the record partitions 1131 a record 1130 ) received by the first node device. As previously discussed, such displays may be considered part of the recurring transmissions of node status 1535 be transmitted to the coordinating device.

at 2520 For example, the processor may delay the assignment of a second task execution using a second record partition of the same record to any node devices of the node device grid including the first node device for a predetermined period of delay to a second node device of the node device grid in which the second node device Record partition is already stored to give a way to become available. at 2530 For example, the processor may check to see if the second node device in which the second data set partition has already been stored became available at the end of the time period of the delay. If the second node device became available within the time period of the delay, then the processor may provide an indication of the assignment to perform the second task with the second data set partition 2532 transmitted to the second node device.

If the second node device at 2530 however, did not become available within the time period of the delay, then the processor can at 2540 check if there was another node device of the node device grid provided with the second record partition to prepare the other node device to serve as a backup node device for a task including the second record partition. If no such other device is present, then the processor may provide an indication of the assignment of the second task to the second record partition 2542 transmitted to the first node device.

If, however, at 2540 If such another node device is present, then the processor can assist in 2550 check if an indication has been received that the other node device is currently available to be assigned a task. If not, then the processor may again provide an indication of the assignment of the second task to the second record partition 2542 transmitted to the first node device. However, if such an indication of the other currently available node device has been received, then the processor may include an indication of the assignment of performing the second task to the second data set partition 2552 transmitted to the other node device.

In various embodiments, each of the processors may be 1150 . 1350 . 1550 . 1650 and 1850 include any of a variety of commercially available processors. Further, one or more of these processors may include multiple processors, a multithreaded processor, a multi-core processor (whether the multiple cores coexist on the same or separate dies), and / or or another type of multiprocessor architecture that connects multiple physically separate processors.

In a particular embodiment, the CPU 1550 each of the one or more node devices 1500 however, be selected to efficiently perform the analysis of multiple instances of job flows at least partially in parallel. By way of example, the CPU 1550 Integrate a single-instruction-multiple-data (SIMD) architecture, integrate multiple processing pipelines, and / or integrate the ability to support multiple simultaneous execution threads per processing pipeline.

Alternatively or additionally, in a particular embodiment, each GPU 1650 the one or more node devices, the at least one of the GPUs 1650 include multithreaded capabilities and / or multiple processor cores to allow parallel execution of tasks. By way of example, the GPU 1650 integrate an internal architecture designed to allow parallel execution of tasks that use a relatively limited set of instructions over hundreds, thousands, tens of thousands or even more threads of execution to support graphics applications that involve relatively high-resolution images.

In various embodiments, each of the control routines may be 1140 . 1340 . 1540 . 1840 and 1640 , including the components of which each is composed, are selected to be operable on any type of processor or processors selected to be applicable to the processors 1150 . 1350 . 1550 . 1850 and or 1650 within the corresponding of the devices 1100 . 1300 . 1500 and / or the graphics controller 1600 to implement. In various embodiments, each of these routines may include one or more of an application-level operating system, device drivers, and / or routines (e.g., so-called "software suites" provided on disc media, "applets" remote from one Server, etc.). If an operating system is included, the operating system may be any of a variety of available operating systems for the processors 1150 . 1550 and or 1850 are suitable. If one or more device drivers are included, these device drivers may support any of a variety of other components, whether hardware or software components, of the devices 1100 . 1300 . 1500 . 1800 and / or 1600.

In various embodiments, each of the memories 1160 . 1360 . 1560 . 1660 and 1860 are based on any of a variety of information storage technologies, including volatile technologies that require the uninterruptible provision of electrical power, and / or technologies that involve the use of machine-readable storage media that may or may not be removable. Thus, each of these memories may include any of a variety of types (or combinations of types) of memory device including, without limitation: read-only memory (ROM), random access memory (RAM), dynamic random access memory (DRAM), double-data-rate DRAM (DRAM), synchronous DRAM (SDRAM), static RAM (SRAM), programmable ROM (PROM), erasable programmable ROM (Erasable Programmable ROM, EPROM) ), Electrically Erasable Programmable ROM (EEPROM), Flash Memory, Polymer Memory (eg, Ferroelectric Polymer Memory), Ovonic Memory, Phase Change or Ferroelectric Memory, Silicon Oxide Nitride Oxide Silicon (SONOS ) Memory, magnetic or optical cards, one or more individual ferromagnetic disk drives, non-volatile storage class memory, or a plurality of memory devices, in one or more reren arrays are organized (z. B. several ferromagnetic disk drives organized in a redundant array of independent hard disks (redundant array of independent disks, RAID). It should be noted that although each of these memories is represented as a single block, one or more of them may include multiple memory devices that may be based on different memory technologies. Thus, for example, one or more of each of the illustrated memories may include a combination of an optical drive or flash memory card reader through which programs and / or data may be stored and communicated on any form of machine-readable storage medium, a ferromagnetic disk drive, programs, and / or Storing data locally for a relatively longer period of time, and one or more volatile solid-state storage devices that provide relatively quick access to programs and / or data (eg, SRAM or DRAM). It should also be noted that each of these memories may consist of multiple memory components based on identical memory technology, but which may be separately routed due to the specialization in use (eg, some DRAM devices are used as a main memory, while other DRAM devices are used as a different frame buffer of a graphics controller).

In a particular embodiment, the memory may be 1160 however, in embodiments in which the one or more of the memory devices 1100 the storage of one or more records 1130 deploy or where the non-volatile memory 1563 in embodiments in which the node devices 1500 the storage of one or more records 1130 deploy with a redundant Array of Independent Disks (RAID) of a RAID level selected to provide fault tolerance for storing one or more records 1130 provide.

In various embodiments, each of the input devices 1320 and 1820 Any of a variety of types of input device may be any of a variety of input detection and / or reception technologies. Examples of such input devices include, but are not limited to, microphones, remote controls, styluses, card readers, fingerprint readers, virtual reality interaction gloves, input graphic tablets, joysticks, keyboards, retinal scanners, touchscreen touch input components, trackballs, environmental sensors, and / or either camera or camera assemblies Monitoring movement of persons to accept commands and / or data provided by these persons through gestures and / or facial expressions.

In various embodiments, each of the displays 1380 and 1880 Any of a variety of types of display device may be any of a variety of visual presentation technologies. Examples of such a display device include, but are not limited to, a cathode ray tube (CRT), an electroluminescent (EL) panel, a liquid crystal display (LCD), a gas plasma display, etc. In some embodiments, the displays can 1180 and or 1880 each be a touch screen display, so that the input devices 1110 and or 1810 each as touch-sensitive components thereof may be integrated therein.

In various embodiments, each of the network interfaces may be 1190 . 1390 . 1590 and 1890 use any of a variety of communication technologies to enable these devices to be coupled to other devices as described. Each of these interfaces includes circuitry that provides at least some of the functionality required to facilitate such coupling. However, each of these interfaces may also be implemented, at least in part, with sequences of instructions executed by corresponding ones of the processors (eg, to implement a protocol stack or other features). If electrically and / or optically conductive cabling is used, these interfaces may use timings and / or protocols that comply with any of a variety of industry standards, including, without limitation, RS-232C, RS-422, USB, Ethernet ( IEEE 802.3 ) or IEEE 1394 , If the use of wireless transmissions is involved, these interfaces may use timings and / or protocols that comply with any of a variety of industry standards, including without limitation: IEEE 802.11a . 802.11ad . 802.11ah . 802.11ax . 802.11b . 802.11g . 802.16 . 802.20 (commonly referred to as "Mobile Broadband Wireless Access");Bluetooth;ZigBee; or a cellular radiotelephone service such as GSM with General Packet Radio Service (GSM / GPRS), CDMA / 1xRTT, Enhanced Data Rates for Global Evolution (EDGE), Evolution Data Only / Optimized (EV-DO), Evolution For Data and Voice (EV -DV), High Speed Downlink Packet Access (HSDPA), High Speed Uplink Packet Access (HSUPA), 4G LTE etc.

In a particular embodiment, one or more of the network interfaces 1190 . 1390 and or 1590 however, with multiple copper-based or fiber optic-based network interface ports implemented to redundant and / or parallel paths when exchanging one or more of the data set partitions 1131 , the task routines 1211 , the CPU task routines 1571 and / or the GPU task routines 1671 provide.

In various embodiments, the allocation of processing and / or storage resources to the node devices 1300 and / or the API architectures used to facilitate communication between the node devices and other devices may be designed and / or selected to comply with any of a variety of distributed processing standards, including but not limited to IEEE P2413 By way of example, a subset of API and / or other architectural features of one or more of such standards may be used to implement the relatively minimal degree of coordination described herein for greater efficiency in parallelizing the processing of data while minimizing an exchange of coordinating information that can lead to unwanted instances of serialization between processes. It should be noted, however, that the parallelization of Storage, retrieval and / or processing of parts of the records 1130 does not depend on or is limited by existing API architectures and / or supporting communication protocols. More generally, there is nothing in the way in which the records are 1130 during storage, transmission and / or distribution over the network 1999 organized to existing API architectures or protocols.

Some systems can use Hadoop ^®, an open-source framework for storing and analyzing big data in a distributed computing environment. Some systems may use cloud computing that can provide ubiquitous, convenient network access on demand to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be deployed quickly with minimal management effort or minimal service provider interaction and can be shared. Some grid systems can be implemented as Hadoop ^® multi-node cluster, as will be understood by those skilled in the art. Apache ^{TM Hadoop®} is an open source software framework for distributed computing.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited non-patent literature

• IEEE 802.3 [0290]
• IEEE-1394 [0290]
• IEEE 802.11a [0290]
• 802.11ad [0290]
• 802.11ah [0290]
• 802.11ax [0290]
• 802.11b [0290]
• 802.11g [0290]
• 802.16 [0290]
• 802.20 [0290]
• IEEE P2413 [0292]

Claims (30)

An apparatus comprising a processor and a memory to store instructions that, when executed by the processor, cause the processor to perform operations comprising: Analyzing a current status of resources from at least one node device of a plurality of node devices to determine availability of at least one graphics processing unit (GPU) of the at least one node device to be assigned to perform a first task of an analysis routine, wherein : the operation of the plurality of node devices is coordinated to perform tasks of analysis routines at least partially in parallel; the analysis routine is generated for execution by at least one central processing unit (CPU) of the at least one node; and the resources of the at least one node device are selected from a group consisting of the at least one CPU, the at least one GPU, and storage space within at least one memory of the at least one node device; in response to a determination that the at least one GPU is available to be assigned to perform the first task of the analysis routine: Analyzing a first task routine of the analysis routine to determine if the first task routine can be compiled to generate a GPU task routine for execution by the at least one GPU to cause the at least one GPU to execute multiple instances of the first task of the analysis routine at least partially in parallel without a dependency between inputs and outputs of the multiple instances of the first task, where: the first task routine is executed for execution by the at least one CPU to perform the first task of the analysis routine; and determining if the first task routine can be compiled to generate the GPU task routine, determining if the first task routine includes an instruction that prohibits compilation to generate the GPU task routine, and determining if inputs and Outputs of the first task routine are defined so as not to require dependency; and in response to a determination that the first task routine can be compiled to generate the GPU task routine: Assigning a record partition of a plurality of record partitions of a record to the at least one node device to allow access to the record partition by the at least one GPU; Using a conversion rule to translate at least one instruction of the first task routine into at least one corresponding instruction of the GPU task routine; Compiling the at least one corresponding instruction of the GPU task routine for execution by the at least one GPU; and Assigning execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the multiple instances of the first task with the record partition by the at least one GPU. The apparatus of claim 1, wherein for determining whether the first task routine includes an instruction that prevents compilation to produce the GPU task routine, the processor is caused to: Determining whether the instruction of the first task routine is included in a set of instructions that can not be converted into at least one instruction that can be executed by the at least one GPU; and in response to a determination that the instruction of the first task routine is not included in the set of instructions, determining whether the instruction of the first task routine in the first task routine is being used in a manner that prevents conversion to at least one instruction being executed by the first task routine at least one GPU can be executed. The apparatus of claim 1, wherein for converting the at least one instruction of the first task routine into the at least one corresponding instruction of the GPU task routine, the processor is instructed to transition the at least one instruction of the first task routine from a first programming language to the at least one corresponding instruction in a second one Programming language according to the conversion rule. Apparatus according to claim 1, wherein: the at least one memory of the at least one node device comprises a first volatile memory communicatively coupled to the at least one CPU and a second volatile memory communicatively coupled to the at least one GPU; assigning the record partition to the at least one node device to allow access to the record partition by the at least one GPU includes causing the record partition to be stored within the second volatile memory; and in response to a determination that the at least one GPU is not available to be assigned to perform the first task of the analysis routine, the processor is caused to perform operations, comprising: omitting the analysis of the first task routine to determine whether the first task routine can be compiled to generate the GPU task routine; Assigning the record partition to the at least one node device to cause storage of the record partition within the first volatile memory to allow access to the record partition by the at least one CPU; Compiling the first task routine for execution by the at least one CPU; and assigning the performing the first task of the analysis routine with the record partition to the at least one node device to enable the first task with the record partition to be performed by the at least one CPU. Apparatus according to claim 1, wherein: the apparatus comprises a coordinator device that coordinates the operation of the plurality of node devices; the processor is caused to periodically receive updates to the current status of each node device of the plurality of node devices; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the processor is caused to identify a node device of the plurality of node devices that integrates a GPU indicated by the current status as available. The apparatus of claim 5, wherein for assigning the record partition of the record to the at least one node device, the processor is caused to perform operations comprising: Analyzing metadata indicative of structural features of the data set to identify a constraint in a manner in which the data set can be divided into the plurality of data set partitions, the constraint being a group consisting of a least atomic unit indication of Data is selected within the record and a specification of a partitioning scheme; and Deriving a division of the record into the plurality of record partitions based at least in part on the constraint. The apparatus of claim 6, wherein the processor is caused to perform operations, comprising: Retrieving the metadata from at least one storage device in which the record is stored; and Transmitting an indication of the allocation of the record partition to the at least one node device or the at least one storage device to cause a transfer of the record partition from the at least one storage device to the at least one node device. Apparatus according to claim 1, wherein: the device comprises a node device of the at least one node device; the node device comprises a GPU of the at least one GPU; the processor includes a CPU that includes at least one CPU; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the CPU is caused to determine whether the GPU of the node device is indicated as being available by the current status. The apparatus of claim 1, wherein the processor is caused to perform operations, comprising: analyzing a second task routine of the analysis routine to determine if the second task routine can be compiled to generate another GPU task routine for execution by the at least one GPU in order to cause the at least one GPU to at least partially parallel multiple instances of the second task of the analysis routine without a dependency between inputs and outputs performing a plurality of instances of the second task, wherein the second task routine is generated for execution by the at least one CPU to perform a second task of the analysis routine; and in response to a determination that the second task routine can not be compiled to generate the other GPU task routine: compiling the second task routine for execution by the at least one CPU; and assigning the second task of the analysis routine with the record partition to the at least one node device to enable the second task with the record partition to be performed by the at least one CPU. The device of claim 1, wherein the conversion rule is selected from a group consisting of: a specification of a set of instructions, each of which prevents compilation of the first task routine to generate the GPU task routine, if present within the first task routine; a specification of a set of instructions, each of which would not prevent the compilation of the first task routine to generate the GPU task routine, if present within the first task routine; specifying a way of using an instruction that prevents compilation of the first task routine to generate the GPU task routine if the manner of using the instruction occurs within the first task routine, otherwise the presence of the instruction within the first task routine does not Prevents compilation of the first task routine for generating the GPU task routine; a specification of a procedure for converting instructions in the first task routine to be looped by the at least one CPU into corresponding instructions of the GPU task routine executed in parallel by the at least one GPU in a corresponding loop with fewer iterations than the one Loop to be executed; a specification of a procedure for converting instructions in the first task routine to be looped by the at least one CPU into corresponding instructions of the GPU task routine to be executed in parallel by the at least one GPU and not in a loop; and a specification of a procedure for transforming instructions in the first task routine that define a data structure comprising entries to be sequentially accessed by the at least one CPU during execution of the first task routine, into corresponding instructions of the GPU task routine that has a corresponding one Defining a data structure comprising entries to be accessed in parallel during execution of the GPU task routine by the at least one GPU. A computer program product concretely embodied in a non-transitory machine-readable storage medium, the computer program product comprising instructions operable to cause a processor to perform operations, comprising: analyzing a current status of resources from at least one node device of a plurality of node devices to determine availability of at least one graphics processing unit (GPU) of the at least one node device to be assigned to perform a first task of an analysis routine, wherein: the operation of the plurality of node devices is coordinated to perform tasks of the analysis routine at least partially in parallel; the analysis routine is generated for execution by at least one central processing unit (CPU) of the at least one node; and the resources of the at least one node device are selected from a group consisting of the at least one CPU, the at least one GPU, and storage space within at least one memory of the at least one node device; in response to a determination that the at least one GPU is available to be assigned to perform the first task of the analysis routine: analyzing a first task routine of the analysis routine to determine whether the first task routine can be compiled to a GPU task routine for execution by the at least one GPU to cause the at least one GPU to perform at least partially in parallel multiple instances of the first task of the analysis routine without a dependency between inputs and outputs of the multiple instances of the first task, wherein: the first task routine for Execution by which at least one CPU is generated to perform the first task of the analysis routine; and determining whether the first task routine can be compiled to produce the GPU task routine, a determination of whether the first task routine includes an instruction that prevents compilation to complete the GPU task routine. Task routine, and determining whether inputs and outputs of the first task routine are defined so as not to require dependency; and in response to a determination that the first task routine can be compiled to generate the GPU task routine: assigning a record partition of a plurality of record partitions of a record to the at least one node device to allow access to the record partition by the at least one GPU ; Using a conversion rule to translate at least one instruction of the first task routine into at least one corresponding instruction of the GPU task routine; Compiling the at least one corresponding instruction of the GPU task routine for execution by the at least one GPU; and assigning execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the plurality of instances of the first task with the record partition by the at least one GPU. The computer program product of claim 11, wherein determining whether the first task routine includes an instruction that prevents compilation to produce the GPU task routine causes the processor to: Determining whether the instruction of the first task routine is included in a set of instructions that can not be converted into at least one instruction that can be executed by the at least one GPU; and in response to a determination that the instruction of the first task routine is not included in the set of instructions, determining whether the instruction of the first task routine in the first task routine is being used in a manner that prevents conversion to at least one instruction being executed by the first task routine at least one GPU can be executed. The computer program product of claim 11, wherein for converting the at least one instruction of the first task routine into the at least one corresponding instruction of the GPU task routine, the processor is instructed to transition the at least one instruction of the first task routine from a first programming language to the at least one corresponding instruction in a second Programming language according to the conversion rule. The computer program product of claim 11, wherein: the at least one memory of the at least one node device comprises a first volatile memory communicatively coupled to the at least one CPU and a second volatile memory communicatively coupled to the at least one GPU; assigning the record partition to the at least one node device to allow access to the record partition by the at least one GPU includes causing the record partition to be stored within the second volatile memory; and in response to a determination that the at least one GPU is not available to be assigned to perform the first task of the analysis routine, the processor is caused to perform operations comprising: Omitting the analysis of the first task routine to determine if the first task routine can be compiled to generate the GPU task routine; Assigning the record partition to the at least one node device to cause storage of the record partition within the first volatile memory to allow access to the record partition by the at least one CPU; Compiling the first task routine for execution by the at least one CPU; and Assigning the execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the first task with the record partition by the at least one CPU. The computer program product of claim 11, wherein: the processor is a component of a coordinator that coordinates the operation of the plurality of node devices; the processor is caused to periodically receive updates to the current status of each node device of the plurality of node devices; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the processor is caused to select a node device of the plurality of Identify node devices that integrate a GPU that is displayed as available in the current state. The computer program product of claim 15, wherein for assigning the record partition of the record to the at least one node device, the processor is caused to perform operations comprising: Analyzing metadata indicative of structural features of the data set to identify a constraint in a manner in which the data set can be divided into the plurality of data set partitions, the constraint being a group consisting of a least atomic unit indication of Data is selected within the record and a specification of a partitioning scheme; and Deriving a division of the record into the plurality of record partitions based at least in part on the constraint. The computer program product of claim 16, wherein the processor is caused to perform operations comprising: Retrieving the metadata from at least one storage device in which the record is stored; and Transmitting an indication of the allocation of the record partition to the at least one node device or the at least one storage device to cause a transfer of the record partition from the at least one storage device to the at least one node device. The computer program product of claim 11, wherein: the processor includes a CPU that includes at least one CPU; the CPU is a component of a node device of the at least one node device; the node device comprises a GPU of the at least one GPU; and to analyze the current status to determine the availability of the at least one GPU of the at least one node device, the CPU is caused to determine whether the GPU of the node device is indicated as being available by the current status. The computer program product of claim 11, wherein the processor is caused to perform operations, comprising: Analyzing a second task routine of the analysis routine to determine if the second task routine can be compiled to generate another GPU task routine for execution by the at least one GPU to cause the at least one GPU to execute multiple instances of the second task Performing an analysis routine at least partially in parallel without a dependency between inputs and outputs of the plurality of instances of the second task, wherein the second task routine is generated for execution by the at least one CPU to perform a second task of the analysis routine; and in response to a determination that the second task routine can not be compiled to produce the other GPU task routine: Compiling the second task routine for execution by the at least one CPU; and Assigning execution of the second task of the analysis routine with the record partition to the at least one node device to enable the execution of the second task with the record partition by the at least one CPU. The computer program product of claim 11, wherein the at least one GPU supports execution of the at least one corresponding instruction of the GPU task routine in parallel over at least one thousand threads of execution. A computer-implemented method, comprising: analyzing a current status of resources of at least one node device of a plurality of node devices to determine availability of at least one graphics processing unit (GPU) of the at least one node device to be assigned a first task performing an analysis routine, wherein: the operation of the plurality of node devices is coordinated to perform tasks of analysis routines at least partially in parallel; the analysis routine is generated for execution by at least one central processing unit (CPU) of the at least one node; and the resources of the at least one node device are selected from a group consisting of the at least one CPU, the at least one GPU, and storage space within at least one memory of the at least one node device; in response to a determination that the at least one GPU is available to be assigned to perform the first task of the analysis routine: analyzing a first task routine of the analysis routine to determine whether the first task routine can be compiled to a GPU task routine for execution by the at least one GPU to cause the at least one GPU to perform at least partially in parallel multiple instances of the first task of the analysis routine without a dependency between inputs and outputs of the multiple instances of the first task, wherein: the first task routine for Execution by which at least one CPU is generated to perform the first task of the analysis routine; and determining whether the first task routine can be compiled to generate the GPU task routine, determining whether the first task routine includes an instruction that prohibits compilation to generate the GPU task routine, and a determination of whether inputs and outputs of the first task routine are defined so as not to require dependency; and in response to a determination that the first task routine can be compiled to generate the GPU task routine: assigning a record partition of a plurality of record partitions of a record to the at least one node device to allow access to the record partition by the at least one GPU ; Using a conversion rule to translate at least one instruction of the first task routine into at least one corresponding instruction of the GPU task routine; Compiling the at least one corresponding instruction of the GPU task routine for execution by the at least one GPU; and assigning execution of the first task of the analysis routine with the record partition to the at least one node device to enable the execution of the plurality of instances of the first task with the record partition by the at least one GPU. The computer-implemented method of claim 21, wherein determining whether the first task routine includes an instruction that prevents compilation to produce the GPU task routine comprises: Determining whether the instruction of the first task routine is included in a set of instructions that can not be converted into at least one instruction that can be executed by the at least one GPU; and in response to a determination that the instruction of the first task routine is not included in the set of instructions, determining whether the instruction of the first task routine in the first task routine is being used in a manner that prevents conversion to at least one instruction being executed by the first task routine at least one GPU can be executed. The computer-implemented method of claim 21, wherein converting the at least one instruction of the first task routine into the at least one corresponding instruction of the GPU task routine comprises transmitting at least one instruction of the first task routine from a first programming language to the at least one corresponding instruction in a second programming language to transform the conversion rule. The computer-implemented method of claim 21, wherein: the at least one memory of the at least one node device comprises a first volatile memory communicatively coupled to the at least one CPU and a second volatile memory communicatively coupled to the at least one GPU; assigning the record partition to the at least one node device to allow access to the record partition by the at least one GPU includes causing the record partition to be stored within the second volatile memory; and in response to a determination that the at least one GPU is not available to be assigned to perform the first task of the analysis routine, the method comprises: omitting the analysis of the first task routine to determine if the first task routine can be compiled to generate the GPU task routine; Assigning the record partition to the at least one node device to cause storage of the record partition within the first volatile memory to allow access to the record partition by the at least one CPU; Compiling the first task routine for execution by the at least one CPU; and assigning the performing the first task of the analysis routine with the record partition to the at least one node device to enable the first task with the record partition to be performed by the at least one CPU. The computer-implemented method of claim 21, wherein: a coordinating device coordinates the operation of the plurality of node devices; the coordinator recurrently receives updates to the current status of each node device of the plurality of node devices; and analyzing the current status to determine the availability of the at least one GPU of the at least one node device, identifying at the coordinating device a node device of the plurality of node devices that integrates a GPU displayed as being available in the current state. The computer-implemented method of claim 25, wherein assigning the record partition of the record to the at least one node device comprises: Analyzing, at the coordinator, metadata indicative of structural features of the data set to identify a constraint in a manner in which the data set can be divided into the plurality of data set partitions, the constraint being a group consisting of a display a smallest atomic unit of data within the data set and a specification of a partitioning scheme is selected; and Deriving a division of the record into the plurality of record partitions based at least in part on the constraint. The computer-implemented method of claim 26, comprising: Retrieving the metadata from at least one storage device in which the record is stored; and Transmitting an indication of the allocation of the record partition to the at least one node device or the at least one storage device to cause a transfer of the record partition from the at least one storage device to the at least one node device. The computer-implemented method of claim 21, wherein: analyzing the current state to determine the availability of the at least one GPU of the at least one node device, analyzing a current status of the node device by a CPU of a node device, whether a GPU of the node device is currently available, wherein: the at least one CPU comprises the CPU; the at least one node device comprises the node device; and the at least one GPU comprises the GPU. The computer-implemented method of claim 21, comprising: Analyzing a second task routine of the analysis routine to determine if the second task routine can be compiled to generate another GPU task routine for execution by the at least one GPU to cause the at least one GPU to execute multiple instances of the second task Performing an analysis routine at least partially in parallel without a dependency between inputs and outputs of the plurality of instances of the second task, wherein the second task routine is generated for execution by the at least one CPU to perform a second task of the analysis routine; and in response to a determination that the second task routine can not be compiled to produce the other GPU task routine: Compiling the second task routine for execution by the at least one CPU; and Assigning execution of the second task of the analysis routine with the record partition to the at least one node device to enable the execution of the second task with the record partitions by the at least one CPU. The computer-implemented method of claim 29, comprising analyzing the analysis routine to determine a sequence of tasks of the analysis routine, the order of tasks comprising a relative order of the first and second tasks.

Images