KR102110027B1 - Transactional memory partition apparatus and method - Google Patents

Abstract

Disclosed is a transactional memory partitioning device and method. The transactional memory partitioning method includes firstly dividing the transactional memory into a plurality of transaction blocks, and processing a transaction in each of the plurality of transaction blocks, wherein the abort transaction block in which processing of the transaction has failed, comprises The second partitioning may include partitioning into a partition HTM (Hardware Transactional Memory), and processing the unsuccessful abort transaction in the plurality of partitions HTM.

Description

TRANSACTIONAL MEMORY PARTITION APPARATUS AND METHOD}

The present invention relates to a transactional memory partitioning apparatus and method for processing large transactions through memory partitioning in a multicore in-memory database environment.

Transactional memory has significantly changed the concurrency control paradigm by replacing the traditional parallel programming technique of lock. Recently, a number of hardware transactional memory (HTM) based studies supported by hardware cache and bus protocols have been conducted.

For example, Intel's Transactional Synchronization Extensions (TSX) is the first HTM technology built into commercial processors, which greatly improves the efficiency of transactional concurrency control.

However, if the tracking information of the transaction exceeds the cache resource limit, a capacity abort may occur, resulting in a transaction failure.

Therefore, there is a need for a technique capable of solving the problem of acceptance failure due to cache resource limitation.

In the present invention, in each of the plurality of transaction blocks partitioning the transactional memory, the transaction is processed, but the transaction block that failed to process the transaction is divided into a plurality of partitions HTM, and the transaction that failed to process the transaction is failed, An object of the present invention is to solve a problem of capacity abort due to a cache resource limit by processing the plurality of partitions HTM.

In order to achieve the above object, the transactional memory partitioning method comprises: firstly dividing the transactional memory into a plurality of transaction blocks; and processing transactions in each of the plurality of transaction blocks, but processing of the transaction fails. It may include the step of dividing the vault transaction block into a plurality of partition HTM (Hardware Transactional Memory), and processing the abort transaction in which the processing has failed, in the plurality of partition HTM.

To achieve the above object, the transactional memory partitioning apparatus includes a control unit for firstly partitioning the transactional memory into a plurality of transaction blocks, and a processor for processing transactions in each of the plurality of transaction blocks, wherein the control unit Is a second partition of the transaction block that failed to process the transaction into a plurality of partitions HTM, and the processor can process the transaction that failed to process the transaction in the plurality of partitions HTM.

According to the present invention, in each of the plurality of transaction blocks partitioning the transactional memory, the transaction is processed, but the transaction block that fails to process the transaction is divided into a plurality of partitions HTM, and the transaction that fails to process the transaction is failed. By processing in the plurality of partitions HTM, it is possible to solve the problem of capacity abort due to cache resource limitation. In this case, the present invention can divide a transaction in which a capacity abort occurs by providing an interface (for example, a Real Time Monitoring (RTM) interface) to which a transaction overlapping technique is applied.

In addition, the present invention can provide an efficient HTM technology in a multicore in-memory environment by providing an adaptive partitioning algorithm to support optimized partitioning according to workload.

The present invention can maximize the utilization of HTM by providing an Eager waiting technique in order to solve the lemming effect that may occur due to Global Lock.

1 is a diagram illustrating an example of a configuration of a transactional memory partitioning device according to an embodiment of the present invention.
2 is a view for explaining an example of transaction processing in a transactional memory partitioning apparatus according to an embodiment of the present invention.
FIG. 3 is a view for explaining an example of transactional memory partitioning and partition path setting in a transactional memory partitioning apparatus according to an embodiment of the present invention.
FIG. 4 is a diagram illustrating an example of pseudocode for transactional memory partitioning in a transactional memory partitioning apparatus according to an embodiment of the present invention.
FIG. 5 is a diagram for explaining an idle waiting technique utilized in a transactional memory partitioning apparatus according to an embodiment of the present invention.
6 is a view for explaining an existing transaction processing technique utilized in an existing transactional memory device.
7 is a flowchart illustrating a transactional memory partitioning method according to an embodiment of the present invention.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings and the contents described in the accompanying drawings, but the present invention is not limited or limited by the embodiments. A transaction processed in the transactional memory partitioning device of the present invention is a group of instructions, and is composed of atomicity, isolation, durability, consistency, and sequentiality. It can have properties.

1 is a diagram illustrating an example of a configuration of a transactional memory partitioning device according to an embodiment of the present invention.

Referring to FIG. 1, the transactional memory partitioning apparatus 100 according to an embodiment of the present invention may include a control unit 101, a confirmation unit 103, and a processor 105.

The control unit 101 may first divide the transactional memory into a plurality of transaction blocks.

In addition, when the processor 105 determines that there is an abort transaction block in which the processing of the transaction has failed, when processing a transaction, the processor 101 may process a plurality of abort transaction blocks in which the processing of the transaction has failed. The partition can be divided into a second into hardware transactional memory (HTM). In this case, the control unit 101 may divide the abort transaction block into a second by the acknowledgment unit 103 when the cause of failure to process the transaction in the abort transaction block is a capacity abort. have. Here, the controller 101 may divide the abort transaction block into a second state in a state in which the transactions are simultaneously processed in each of the plurality of transaction blocks, excluding the abort transaction block.

In the second partition, the controller 101 determines the number of partition HTMs to be partitioned based on the write-set size for the abort transaction or the capacity of the transactional memory. , The abort transaction block may be divided into a second number by the determined number. At this time, the controller 101 may divide the transactional memory evenly, but is not limited thereto, and may be divided into various sizes (capacity).

In addition, the controller 101 may adjust the capacity of the second partition HTM based on the write size. At this time, the control unit 101 may adjust the capacity of the partition HTM in proportion to the write size, but when the write size is less than or equal to a first set value or a second set value (greater than the first set value), the second The capacity of the partition HTM to be divided can be adjusted to a reference capacity (a specific predetermined capacity).

When it is determined by the processor 105 that there is an abort transaction block in which the processing of the transaction has failed, the confirmation unit 103 may check the cause of the failure to process the transaction in the abort transaction block.

The processor 105 may process a transaction in each of a plurality of divided transaction blocks by the control unit 101.

In addition, as the processor 105 determines that there is an abort transaction block in which the processing of the transaction has failed, the abort transaction block in which the processing of the transaction has failed is processed by the control unit 101 into a plurality of partitions HTM. In the case of partitioning, the abort transaction in which the processing has failed can be processed in the partitioned plurality of partitions HTM. At this time, when the processor 105 processes the abort transaction in any one partition HTM among the plurality of partition HTMs, the processor 105 may stop processing of the abort transaction in the remaining partition HTM.

In the plurality of partitions HTM, when the processing of the abort transaction fails, the processor 105 may give a lock command to the abort transaction and maintain it in a predetermined memory area.

On the other hand, if the cause of the failure to process the transaction is not an accommodating failure, the processor 105 reprocesses the abort transaction in the abort transaction block, and if the number of reprocessing exceeds a predetermined number, the abort transaction By giving a lock command to, it can be held in a designated memory area. At this time, the processor 105, even if the number of reprocessing does not exceed the predetermined number, through the interface unit (not shown), when receiving a lock command, grants a lock command for the abort transaction, It can be kept in a designated memory area.

After processing the transaction through the transaction block or the partition HTM, the processor 105 may issue a complete command to the abort transaction maintained in the memory area to process the abort transaction.

2 is a view for explaining an example of transaction processing in a transactional memory partitioning apparatus according to an embodiment of the present invention.

Referring to FIG. 2, in step 201, the transactional memory partitioning device may start a Lite HTM utilizing the existing HTM as it is at the start of the transaction block.

In step 202, the transactional memory partitioning device may restart a transaction that failed due to a capacity abort during transaction processing with the Lite HTM to a partitioned HTM.

In step 203, the transactional memory partitioning device performs a certain retry to ensure maximum success for transactions that fail for reasons other than acceptance failure, and in step 204, for a transaction that repeats the failure, the global lock ( Global Lock) to ensure the completion of the serial command (commit).

In step 205, the transactional memory partitioning device may split a transaction based on an appropriate write size by calculating a partitioning criterion reflecting workload characteristics in a split manager for a transaction restarted in the partition HTM.

In step 206, the transactional memory partitioning device performs a certain retry to ensure maximum success for the partition HTM, as in the Lite HTM, and in step 207, until the commit part of the global transaction is met. The division processing can be repeated.

In step 208, the transactional memory partitioning device may acquire a global lock and guarantee a completion command serially in the same manner as the Lite HTM for the partition HTM that failed despite the partitioning process.

In step 209, the transactional memory partitioning device may collect write-set information through execution of a serial processing transaction in which a global lock is acquired. At this time, the transactional memory partitioning device simultaneously synchronizes both the Lite HTM and the Partition HTM that do not acquire the Global Lock through the MESI (Modified Exclusive Shard Invalid) protocol, which is one of the cache coherency protocols as the existing HTM. Control can be guaranteed.

FIG. 3 is a view for explaining an example of transactional memory partitioning and partition path setting in a transactional memory partitioning apparatus according to an embodiment of the present invention.

Referring to FIG. 3, the transactional memory partitioning apparatus may perform partitioning while maintaining execution of a global transaction in a partitioned path. At this time, the transactional memory partitioning device maintains the HTM of the global transaction by supporting the overlap by maintaining the number of calls of the _xbegin () and _xend () commands specified by the programmer in memory through the Real Time Monitoring (RTM) interface. And, you can easily set the HTM range of the partition (partition) transaction.

 Here, the transactional memory partitioning device prevents a transaction from being performed between one partition path and the next partition HTM connected to ensure the isolation of the transaction. In addition, the transactional memory partitioning device can guarantee accuracy by performing a rollback to the immediately preceding _xbegin () through the _xabort () command when the transaction fails.

Specifically, the transactional memory partitioning device may perform global transaction start with HTM in the same way as normal transaction processing at the starting point of a transaction block. Thereafter, the transactional memory partitioning device may repeat the completion command and the start command every time the partition threshold is reached until all transactions that are performed encounter the global transaction's commit command. . That is, the transactional memory partitioning apparatus may perform an HTM start command (begin) and a completion command (commit) in an overlapped state while the HTM of the global transaction is maintained.

In addition, the transactional memory partition unit partitions a partition command to match a pair of a start command and a commit command even when a transaction block that has not reached the partition threshold reaches the end of the global transaction. You can do

On the other hand, the transactional memory partitioning device can calculate through the number of write transactions that have been performed when calculating the transaction block partitioning criteria, but accessing the internal buffer on the HTM to check the write history does not guarantee transaction safety. It can be calculated in a serial processing environment where a global lock is acquired. The transactional memory partitioning apparatus can set the numerical value α divided by the variable γ to the total write size (write-set) as a partitioning threshold under the assumption that the write-set of the existing transaction block is a burden on the cache. . The transactional memory partitioning apparatus can calculate the factor γ, which sets α, in cycles of 100 transactions in order to fully reflect the workload characteristics in real time. Here, γ is the number of successive successive commits based on the existing α among 100 transaction blocks, and may be reduced to expand the division criterion when β is 50 or more. Otherwise, the transactional memory partitioning apparatus may increase γ by determining that it is necessary to refine the transaction partitioning criteria.

The transactional memory partitioning device does not reflect the workload in real time, and when providing a static partitioning criterion, the value of γ may be limited to, for example, 5 and a positive integer less than 10. At this time, the transactional memory partitioning device may implement this in pseudocode, for example, as shown in FIG. 4.

4 is a diagram illustrating an example of pseudocode for transactional memory partitioning in a transactional memory partitioning apparatus according to an embodiment of the present invention.

Referring to FIG. 4, the transactional memory partition apparatus sets a period variable according to the occurrence of capacity abort in order to set, for example, 100 execution times of a transaction through a partition path. Can be increased (line1-2) 401.

In addition, the transactional memory partitioning device performs a sequential commit when a partitioned transaction block is committed, in order to measure the number of commits β of a continuous partition path. It increases, and if it fails, it can be initialized (line3-6) (403).

The transactional memory partitioning apparatus may periodically calculate a partitioning criterion (γ) every time the period becomes 100, and reflect it in a partitioning threshold, that is, a split size (α). At this time, the transactional memory partition apparatus may calculate the split size based on the number of write instructions.

FIG. 5 is a diagram illustrating an Eager waiting technique utilized in a transactional memory partitioning apparatus according to an embodiment of the present invention.

Referring to FIG. 5, the transactional memory partitioning device may support an awaiting mechanism that waits according to whether a global mutex lock occurs immediately after a transaction block is called.

That is, the transactional memory partitioning device does not attempt unnecessary execution that may cause an abort by the global mutex lock when processing a transaction on the HTM, and the global mutex lock Efficient retries in the free state of) can solve the problem of the remming effect that leads to a chain of abort and fallback paths.

The transactional memory partitioning device is a method of waiting according to whether or not a global mutex lock is present. By performing a pause through a system function, it can be paused for input of an internal buffer. In comparison, energy consumption can also be reduced.

As a result, the transactional memory partitioning device not only supports efficient transaction processing by utilizing the cache and bus protocols, but also solves the problem of the reaming effect that may occur due to the application of the existing transaction processing technique shown in FIG. 6 through the weighting technique. Can be solved.

That is, in a multi-core environment, a transactional memory device that employs an existing transaction processing technique, which consists of two or more paths, including a fallback path, causes transactions processed in the HTM environment to commit. Although there is no problem in execution, as the core 2 global mutex lock consumes the number of retries and aborts it, the forward progress of the transaction performed in Core 2 However, while reducing the HTM's commit rate performed by the remaining cores, multiple transactions may cause a problem with the reaming effect leading to a fallback path. On the other hand, the transactional memory partitioning device of the present invention solves the reaming effect problem through a weighting technique, thereby guaranteeing forward progress, and implements concurrency control technology even in two or more paths. Can be.

7 is a flowchart illustrating a transactional memory partitioning method according to an embodiment of the present invention.

Referring to FIG. 7, in step 701, the transactional memory partitioning device may first divide the transactional memory into a plurality of transaction blocks.

In step 703, the transactional memory partitioning device may process a transaction in each of the plurality of transaction blocks.

In step 705, the transactional memory partitioning device determines whether there is an abort transaction block in which the processing of the transaction has failed, and if there is an abort transaction block in which the processing of the transaction has failed, in step 707, the transaction is processed. It is possible to check the cause of the failure.

As a result of the check, in step 709, if the cause of failure to process the transaction is an acceptance failure, in step 709, in step 711, the transaction block of the transaction that failed to process the transaction is a plurality of partitions HTM Can be divided into the second. At this time, the transactional memory partitioning device may divide the abort transaction block into a second state in a state in which the transactions are simultaneously processed in each of the plurality of transaction blocks, excluding the abort transaction block.

Here, the transactional memory partitioning device determines the number of the partition HTM to be partitioned, based on the write size of the transaction or the capacity of the transactional memory, and determines the number of the transaction The block can be divided into second.

In addition, the transactional memory partition apparatus may adjust the capacity of the second partition HTM based on the write size.

In step 713, the transactional memory partitioning apparatus may process the abort transaction in which the processing has failed, in the plurality of partitions HTM. At this time, the transactional memory partitioning device may stop processing of the abort transaction in the remaining partition HTM when the abolt transaction is processed by any one partition HTM among the plurality of partition HTMs.

In addition, the transactional memory partitioning apparatus can maintain a fixed memory area by issuing a lock command to the abort transaction even when the processing of the abort transaction fails in the plurality of partition HTMs.

On the other hand, as a result of the check in step 707, if the cause of the failure to process the transaction is not an acceptance failure, in step 715, the transactional memory partitioning device reprocesses the abort transaction in the abort transaction block. can do. At this time, if the number of times the reprocessing exceeds a predetermined number, the transactional memory partitioning device may grant a lock command to the abort transaction and maintain it in a predetermined memory area.

After processing the transaction through the transaction block or the partition HTM, the transactional memory partitioning device may process the abort transaction by giving a complete command to the abort transaction maintained in the memory area.

The device described above may be implemented with hardware components, software components, and / or combinations of hardware components and software components. For example, the devices and components described in the embodiments include, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors (micro signal processors), microcomputers, field programmable arrays (FPAs), It may be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may run an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and / or a plurality of types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of these, and configure the processing device to operate as desired, or process independently or collectively You can command the device. Software and / or data may be interpreted by a processing device, or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodied in the transmitted signal wave. The software may be distributed over networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable storage media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and stored in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions stored in the medium may be specially designed and configured for an embodiment or may be known and available to those skilled in computer software. Examples of computer-readable storage media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. Includes hardware devices specifically configured to store and execute program instructions such as magneto-optical media and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described by a limited embodiment and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and / or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

100: transactional memory partition device
101: control
103: confirmation unit
105: processor

Claims (12)

First dividing the transactional memory into a plurality of transaction blocks;
Processing a transaction in each of the plurality of transaction blocks;
Confirming a cause of the failure to process the transaction in the abort transaction block in which the processing of the transaction has failed;
Dividing the abort transaction block into a plurality of partition hardware transactional memories (HTMs) when the identified cause is a capacity abort;
Adjusting the capacity of the second partitioned partition HTM based on a write-set size for the abort transaction in which the processing has failed; And
As the step of processing the abort transaction in the plurality of partition HTMs, when the abort transaction is processed in any one partition HTM among the plurality of adjusted partition HTMs having various sizes, the remaining partitions HTM are processed. Stopping processing of the abort transaction
Transactional memory partitioning method comprising a. According to claim 1,
Even in the plurality of partitions HTM, when the processing of the abort transaction fails,
The step of granting a lock command to the abort transaction and maintaining it in a predetermined memory area.
A transactional memory partitioning method further comprising a. According to claim 1,
If the above identified cause is not an accommodation disorder,
Reprocessing the abort transaction in the abort transaction block; And
When the number of times of reprocessing exceeds a predetermined number, a lock command is given to the abort transaction, and the result is maintained in a predetermined memory area.
A transactional memory partitioning method further comprising a. The method of claim 2 or 3,
Processing the abort transaction by giving a completion command to the abort transaction maintained in the memory area
A transactional memory partitioning method further comprising a. According to claim 1,
Determining the number of the second partitioned partition HTM based on the capacity of the transactional memory
Transactional memory partitioning method comprising a. delete delete According to claim 1,
The second dividing step,
The second step of dividing the abort transaction block in a state in which the transactions are simultaneously processed in each of the plurality of transaction blocks except for the abort transaction block.
Transactional memory partitioning method comprising a. delete A control unit for first dividing the transactional memory into a plurality of transaction blocks;
A processor that processes a transaction in each of the plurality of transaction blocks; And
Confirmation unit that checks the cause of the failure to process the transaction in the Abort transaction block where the transaction has failed.
Including,
The control unit,
When the identified cause is a handicapped person, the partition of the partition HTM that divides the abort transaction block into a plurality of partition HTMs based on the write size of the abolt transaction in which the processing fails is divided into two. Adjust the capacity,
The processor,
When processing the abort transaction in the plurality of partition HTMs, but processing the abort transaction in any one partition HTM among the plurality of partitions HTMs having the adjusted and varied sizes, the remaining partitions HTM process the abort transaction. To stop processing for a vault transaction
Transactional memory partition. The method of claim 10,
Even in the plurality of partitions HTM, when the processing of the abort transaction fails,
The processor,
A lock command is given to the abort transaction, which is maintained in a predetermined memory area.
Transactional memory partition. The method of claim 10,
If the above identified cause is not an accommodation disorder,
The processor,
Reprocessing the abort transaction in the abort transaction block, and when the number of reprocessing exceeds a predetermined number, a lock command is given to the abort transaction and maintained in a predetermined memory area.
Transactional memory partition.

Images