WO2020034646A1

WO2020034646A1 - Resource scheduling method and device

Info

Publication number: WO2020034646A1
Application number: PCT/CN2019/081200
Authority: WO
Inventors: 刘志飘; 邓慧财; 杰森•T•S• 兰
Original assignee: 华为技术有限公司
Priority date: 2018-08-17
Filing date: 2019-04-03
Publication date: 2020-02-20
Also published as: CN109375992A

Abstract

A resource scheduling method and device, the method specifically comprising: a resource scheduling device receives a task request sent by a client, the task request comprising a resource tag corresponding to the task request, and the task request specifying resources required for executing a task by means of the resource tag comprised in the task request; the resource scheduling device determines a target node from a node group according to resource tags of each node in the node group, the target node having the resource tag comprised in the task request; and the resource scheduling device instructs the target node to execute the task specified in the task request. On one hand, because of the resources required for the task which are specified by the resource tag corresponding to the task request, it is possible to determine a target node of which the processing capabilities are suitable for the task, thereby contributing to an improvement in the efficiency of executing a task specified by a task node. On the other hand, determining a target node having the resource tag comprised in the task request from the node group may help to improve the resource utilization rate of the nodes.

Description

Method and device for resource scheduling

Technical field

The present application relates to the technical field of cloud computing, and in particular, to a method and a device for resource scheduling.

Background technique

In video cloud computing scenarios, such as traffic video, video encoding and decoding, facial feature extraction, license plate recognition, and congestion event monitoring and other processing, it is necessary to implement the tasks that need to be processed to the nodes in the cloud server cluster for processing. Another resource coordinator (Yet, Other, Resource, Negotiator, YARN) is a Hadoop 2.X general-purpose resource management system, responsible for resource management and task scheduling of nodes in a cloud server cluster.

In the prior art, a resource scheduling scheme based on YARN submits a task to a resource queue, and the division of the resource queue is the division of the logical cluster, that is, the task is scheduled to the logical cluster corresponding to the resource queue. On the node. For example, for example, divide the resources of the entire cluster into three resource queues: resource queue 1, resource queue 2, and resource queue 3. Resource queue 1 corresponds to logical cluster A, resource queue 2 corresponds to logical cluster B, and resource queue 3. Corresponding to logical cluster C, if a task 1 is submitted to resource queue 2, the task 1 will be scheduled to a node in logical cluster B. In the resource scheduling scheme in the prior art, a resource queue is bound to a node in a logical cluster. If a resource queue is determined, a node can only be determined in the logical cluster corresponding to the resource queue. It can be seen that in the prior art resource scheduling schemes, tasks may be scheduled to nodes with weak processing capabilities, resulting in low task efficiency and low resource utilization of nodes in the cluster.

In summary, how to improve the execution efficiency of the task and the resource utilization of the nodes still needs further research.

Summary of the Invention

The present application provides a resource scheduling method and device, which are used to improve task execution efficiency and node resource utilization.

In a first aspect, the present application provides a resource scheduling method. The method includes: a resource scheduling device receives a task request sent by a client, the task request includes a resource tag corresponding to the task request, and the task request passes the task. The resource tag included in the request specifies the resources required to perform the task; the resource scheduling device determines a target node from the node group according to the resource tag that each node in the node group has, and the target node has the resources included in the task request Label; the resource scheduling device instructs the target node to execute the task specified by the task request.

Based on this solution, the resource scheduling device receives the task request sent by the client, and the resource scheduling device determines a target node from the node group according to the resource label that each node in the node group has and the resource label included in the task request, and the target The node has a resource tag included in the task request; the resource scheduling device instructs the target node to execute a task specified by the task request. On the one hand, since the task request includes a resource tag corresponding to the task request, the task request specifies a resource required to execute the task through the resource tag included in the task request, so the resource required by the task specified by the task request To determine the target node, a node with a processing capacity suitable for the task specified by the task request can be found, thereby helping to improve the execution efficiency of the task specified by the task node. On the other hand, when determining the target node to execute the task specified by the task request, the resource label included in the task request is determined from the node group according to the resource label that the node has. In this way, the Resources, which in turn helps improve the resource utilization of nodes.

When a resource scheduling device receives a task request and determines a target node from a node group, since some tasks require resources to execute when they include a graphics processing unit (GPU), some tasks require resources that are not required for execution. Including the GPU. Therefore, if a task without a GPU requirement is scheduled to be executed on a node with a GPU, a task that urgently needs GPU resources cannot find a suitable node with a GPU resource, thereby causing a waste of GPU resources on the node. Therefore, in order to avoid wasting GPU resources in the node group, based on the above method, several alternative implementation methods are provided below.

In the first implementation method, the resource scheduling device may select a target node for executing a task specified by the task request according to whether a resource tag corresponding to the task request includes a resource tag of the GPU.

As a specific example, when the resource tag corresponding to the task request does not include the resource tag of the GPU, a node without a resource tag of the GPU is selected from the node group; the selected node is instructed to execute the task specified by the task request.

As another specific example, when a resource tag corresponding to the task request includes a resource tag of a GPU, a node having a resource tag of the GPU is selected from a node group; and the selected node is instructed to execute a task specified by the task request.

With this implementation method, according to whether a resource tag corresponding to a task request includes a GPU tag, selecting a target node to execute a task specified by the task request, it is possible to implement GPU resource scheduling on demand according to the resources required by the task specified by the task request. , Which can improve the resource utilization of the nodes in the node group, especially the utilization of GPU resources, which helps to avoid the exhaustion of the central processing unit (CPU) resources in the node and the Excessive GPU resources are wasted.

In the second method, the task request may further include the amount of resources required to execute the task, and the resource scheduling device may select a target node to execute the task specified by the task request according to whether the amount of resources required to execute the task includes the amount of GPU resources.

As a specific example, if the amount of resources required to execute a task does not include GPU resources, selecting a node that does not have the GPU resources from the node group; instructing the selected node to execute the task specified by the task request task.

As another specific example, if the amount of resources required to perform a task does not include GPU resources, selecting a node that does not have the GPU resources from the node group; instructing the selected node to execute the task request specification Task.

Further, as a possible implementation manner, the task request records the amount of resources required to execute the task; the resource scheduling device determines the target node from the node group according to the resource label that each node in the node group has, and specifically includes: The node group selects a target node that satisfies the amount of resources specified by the task request. Specifically, the resource scheduling device selects, from the node group, a node that satisfies the amount of resources specified by the task request and has the resource tag included in the task request, that is, the target node. In this way, the determined target node not only meets the resources required to execute the task specified by the resource tag included in the task request, but also meets the amount of resources required to execute the task, thereby helping to improve the execution efficiency of the task specified by the task request.

Based on any of the foregoing embodiments, the method may further include: the resource scheduling device obtains resource information of each node in the node group, and the resource information records a node identifier of the node and resources owned by the node; The resource information of each node in the node group is described, and the resource tag and the amount of resources that each node in the node group has are recorded. In this way, the resource scheduling device can accurately determine the node used to perform the task specified in the task request according to the resource label and the amount of resources the node has.

Based on any of the above embodiments, it may further include: when the task specified by the task request has affinity with another task, the resource scheduling device instructs the target node to perform another task; and when the task specified by the task request has another task In the case of anti-affinity, the resource scheduling device instructs another node in the node group different from the target node to perform another task. In this way, when determining a node that executes a task, the resource scheduling device considers the affinity between the tasks, so that a target node suitable for performing the task can be determined.

In a second aspect, the present application provides a resource scheduling method. The method includes: a client generates a task request, the task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to execute the task through the resource tag included in the task request. ; Then, the client sends a task request to the resource scheduling device, and the task request is used by the resource scheduling device to determine the target node to perform the task specified by the task request according to the resource tag of each node in the node group, and the target node has the task request Resource tags.

Based on this solution, the client generates a task request. The task request specifies the resources required to perform the task through the resource tag included in the task request. The client sends a task request to the resource scheduling device. On the one hand, the resource scheduling device specifies the task specified by the task request The required resources are used to determine the target node, and a node with a processing capacity suitable for the task specified by the task request can be found, thereby helping to improve the execution efficiency of the task specified by the task node. On the other hand, when the resource scheduling device determines a target node that executes the task specified by the task request, it determines from the node group that the resource label includes the resource label included in the task request according to the resource label that the node has, so that the node can be used reasonably The resources of the nodes in the group, which in turn help improve the resource utilization of the nodes.

In a third aspect, the present application provides a device having the functions of implementing the embodiments of the first aspect described above. This function can be realized by hardware, and can also be implemented by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions.

In a possible design, the device includes a communication unit and a processing unit, where:

A receiving unit, configured to receive a task request sent by a client, where the task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to perform a task through the resource tag included in the task request;

The processing unit is configured to determine a target node from the node group according to a resource label that each node in the node group has, and the target node has the resource label included in the task request; and instruct the target node to execute all resources. The task specified by the task request.

In a fourth aspect, the present application provides a device having the functions of implementing the embodiments of the second aspect described above. This function can be realized by hardware, and can also be implemented by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions.

A processing unit, configured to generate a task request, where the task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to perform a task by using the resource tag included in the task request;

A communication unit, configured to send the task request to a resource scheduling device, where the task request is used by the resource scheduling device to determine a target node to perform a task specified by the task request according to a resource tag possessed by each node in the node group , The target node has a resource tag included in the task request.

In a fifth aspect, the present application provides a device including: a processor and a memory; the memory is configured to store computer instructions; the processor executes the computer instructions stored in the memory, so that the device executes the foregoing first aspect or the first aspect The resource scheduling method in the method provided by any embodiment. It should be noted that the memory may be integrated in the processor, or may be independent of the processor.

According to a sixth aspect, the present application provides a device including: a processor and a memory; the memory is configured to store computer instructions; the processor executes the computer instructions stored in the memory, so that the device executes the second aspect or the second aspect described above The resource scheduling method in the method provided by any embodiment of the present invention. It should be noted that the memory may be integrated in the processor, or may be independent of the processor.

In a seventh aspect, the present application further provides a system including a resource scheduling apparatus in the method provided by any embodiment of the first aspect and a client in the method provided by any embodiment of the second aspect.

In an eighth aspect, the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores computer instructions, and the computer instructions are executed by a computer to implement the methods described in the foregoing aspects.

In a ninth aspect, the present application also provides a computer program product containing instructions that, when run on a computer, causes the computer to execute the methods described in the above aspects.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1a is a schematic diagram of a possible system architecture provided by this application; FIG.

FIG. 1b is a schematic diagram of another possible system architecture provided by this application; FIG.

FIG. 2 is a schematic diagram of a resource scheduling method provided by the present application; FIG.

FIG. 3 is a schematic diagram of a mapping relationship between a node identifier and a resource tag provided in this application;

4 is a schematic diagram of another resource scheduling method provided by the present application;

FIG. 5 is a schematic structural diagram of a device provided by this application;

FIG. 6 is a schematic structural diagram of another device provided by the present application.

detailed description

The present application will be further described in detail below with reference to the accompanying drawings. The specific operation method in the method embodiment can also be applied to the device embodiment or the system embodiment. Wherein, in the description of the present application, unless otherwise stated, the meaning of "a plurality" is two or more.

FIG. 1a exemplarily illustrates a possible system architecture diagram provided by an embodiment of the present application. The system architecture includes N nodes included in a client, a resource scheduling device, and a node group. FIG. 1a only shows nodes 1, ..., and N by way of example, where N is a positive integer.

The client generates a task request according to the job submitted by the user, and sends the task request to the resource scheduling device. The task request specifies the resources required to perform the task through the resource tag included in the task request. The task request is used for the resource scheduling device slave node. Determines the target node in the group to perform the task specified by the task request.

The resource scheduling device receives the task request sent by the client, and determines the target node for executing the task request from the node group according to the task request and the resources possessed by the N nodes included in the node group, and instructs the target node to execute the task request designation. Task.

The N nodes included in the node group are used to perform tasks instructed by the resource scheduling device to execute.

Based on FIG. 1a and FIG. 1b, another schematic diagram of a possible system architecture provided by an embodiment of the present application is exemplarily shown.

Referring to FIG. 1b, there are exemplarily shown five modules including a task request forwarding module, a task scheduling control module, a label customization module, a scheduling engine (Yarn) module, and a Docker mirror registry (Docker Registry) included in the resource scheduling apparatus.

Among them, the task request forwarding module is responsible for receiving algorithm scheduling tasks of users, providing load balancing and request forwarding capabilities.

The task scheduling control module is used to complete functions such as task parameter analysis, task scheduling control, and task queue management. As an example, the task parameter analysis function may be to analyze the task priority, the task queue management function may be to place the task that has parsed the task parameters into the task queue according to the time sequence of the received task, and the task scheduling control function may be based on the task. The priority and the time when the task is placed in the task queue are scheduled from the task queue to Yarn for processing. Among them, the high priority task is scheduled first, and the tasks with the same priority are scheduled according to the time order of the task queue.

The label customization module mainly includes the information collection and reporting sub-module and the multi-label customization sub-module. The information collection and reporting sub-module is used to collect the resource information of each node in the node group. On the one hand, the information collection and reporting sub-module reports the collected resource information to Yarn, so that Yarn can be used for task scheduling according to the resource information of the node. The resources required to perform this task; on the other hand, the information collection and reporting sub-module sends the collected resource information to the multi-label custom sub-module, so that the label custom sub-module generates resource tags for each type of resource according to the resource information. Report the correspondence between the node identifier and all resource tags that the node has to Yarn.

A scheduling engine (Yarn) module is used to store the correspondence between the received node identifiers and all resource tags possessed by the nodes in a node tag database, and implement specific task scheduling functions according to all resource tags possessed by the nodes. The node tag database is located in Yarn. For example, the node tag database may exist in Yarn in the form of a configuration file.

As a specific example, when Yarn receives a task scheduled by the task scheduling control module, Yarn queries the resource label of each node from the node label resource library, and determines according to the resource label of the node and the resource label of the task request. The node that executes the task specified by the task request.

Docker Registry, which manages open source components for Docker's private image repository, mainly provides image upload and download capabilities.

The resource scheduling method provided in the present application will be described in detail below with reference to the system architecture shown in FIG. 1a and FIG. 1b.

FIG. 2 exemplarily illustrates a schematic flowchart of a resource scheduling method provided by the present application. The method includes the following steps:

Step 201: The client generates a task request.

The task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to execute the task by using the resource tag included in the task request.

Here, the task request may include one or more resource tags. As an implementation manner, resource tags may exist in the task request in the form of a label list.

As an implementation manner, the resource tag corresponding to the task request is determined according to the amount of resources required by the task. Specifically, the client determines the resource tag corresponding to the task request according to the amount of resources required for the task, or the client determines the resource tag corresponding to the task request according to the amount of resources required for the task and a preset rule, and the preset rule is related to The resource scheduling device determines that the rules of the resource tags corresponding to the task requests are consistent.

As another implementation manner, the resource label corresponding to the task request may be obtained from a resource scheduling device. Specifically, the client sends a query message to the resource scheduling device, where the query message is used to obtain a resource label set, and the resource label set includes a resource label of each node in the node group. As an example, the resource label indication of the node The resources the node has. The client then receives a response message from the resource scheduling device, the response message including the resource tag set. After the client obtains the resource tag set, the client can select one or more resource tags from the resource tag set according to the resources required by the task, and carry the selected resource tag to the task request.

As a specific example, the above task request may further include some or all of the following information: task name (name), job identification (identity), algorithm image identification (identity), task priority (priority), copy Number (copy number), the resources required by the task (such as CPU, memory, and GPU resources), and the amount of resources required by the task.

For example, a job may include multiple tasks. A user submits a job as an example of checking violations in a traffic video. The client may generate one or more task requests based on the job. A task request may correspond to a task. Can correspond to multiple tasks. For example, one task request corresponds to one task. For example, the job is divided into two tasks. Task 1 corresponding to task request 1 is to convert a video into an image. Task 2 corresponding to task request 2 is image processing, such as face recognition. In this example, the job identifiers of task 1 and task 2 belonging to the same job are the same, that is, the job identifiers included in task request 1 and the job identifiers included in task request 2 are the same.

In the above example, task 1 and task 2 belong to the same job. Because the tasks in the same job have a dependency relationship, the task priorities of multiple tasks that belong to the same job can be set to the same, that is, tasks included in task request 1. The task priority is the same as the task priority included in task request 2.

Before any task in task 1 or task 2 is scheduled to be executed on a node, a container needs to be generated in the node, and the container includes resources required to perform the task. Taking task 1 to node 1 as an example, the task request 1 indicates the algorithm image identity (image identity) corresponding to the task 1, and Docker Registry pushes the image identified by the algorithm image identity to node 1 to generate a container in node 1.

Step 202: The client sends the task request to the resource scheduling device. Accordingly, the resource scheduling apparatus may receive the task request sent by the client.

In step 203, the resource scheduling device determines a target node from the node group according to a resource label possessed by each node in the node group.

The target node has a resource tag included in the task request.

Specifically, after receiving the task request, the resource scheduling device determines a target node having the resource label included in the task request from the node group according to the resource label corresponding to the task request and the resource label possessed by the node.

Optionally, each node in the node group may have one or more resource tags.

As an implementation manner, the resource scheduling device acquires resource information of each node in the node group, and the resource information records a node identifier of the node and resources possessed by the node; the resource scheduling device according to the resource information of each node in the node group, Record the resource label and the amount of resources each node in the node group has. In this way, the resource scheduling device can accurately determine the node used to perform the task specified in the task request according to the resource label and the amount of resources the node has.

Based on the above-mentioned FIG. 1b, a label customization module may be used to collect resource information of each node in the node group, and generate a resource label that each node has. The node identifier recorded in the resource information may be the IP address of the node, and the resources recorded by the node may be resources such as CPU, GPU, memory, and disk. Optionally, the resource tags possessed by the node may be stored in a node tag database. When the resource scheduling device executes the above step 203, the resource tags of each node are traversed from the node tag database in order to select a target node for the task specified by the task request according to the resource tags of the node.

As an example, as shown in FIG. 3, it is a schematic diagram of a mapping relationship between a node identifier and a resource label. A node may correspond to one or more resource tags, and each resource tag may correspond to one or more nodes. Referring to FIG. 3, taking a node identifier as an internet protocol (IP) address for interconnection between networks as an example, the IP address of node 1 (Node1): 192.168.0.10, corresponding to resource label 1 (Label 1) and Label 3; Node 2's IP address: 192.168.0.11, corresponding to Resource Label 2 (Label 2); Node 3 (Node 3) 's IP address: 192.168.0.12, corresponding to Label 2 and resource label M (Label M); node The IP address of N (Node) is 192.168.0.xx, corresponding to Label 1, Resource Label 3 (Label 3), and Label M, where M is a positive integer.

This example uses the unique identifier of a node (such as an IP address) to map resource tags, and supports a single node to customize multiple resource tags. Compared with the prior art, only native Yarn interface capabilities can be used to indirectly provide nodes with task queues. The scheme of tagging resource tags and the node does not support multiple resource tags. This application can implement customizing multiple resource tags for a node and directly specifying the resource tags corresponding to the task request when the task is issued, which can optimize the resource scheduling process. This will help to improve the utilization of node group resources.

Further, the target node has a resource tag included in the task request, and specifically, there can be at least two implementation manners.

As an implementation manner, the resource tag that the target node has includes a resource tag corresponding to the task request.

Specifically, when the resource scheduling device receives the task request, it is determined from the node group that the node having the resource label including the resource label corresponding to the task request is the target node.

In this implementation, a node may have one or more resource tags corresponding to the following resource types: CPU frequency, GPU model, GPU block number, memory module model, bandwidth, and computer room information. For example, the GPU model can include P4, P40, P100, and V100. For example, the memory module model can include double data rate (DDR) 1, DDR2, DDR4, and so on. Exemplarily, for example, node 1 has one resource label, which indicates that the CPU clock frequency is high; for another example, node 2 has two resource labels, which indicate the CPU clock frequency is low and GPUP100 block. The resource labels corresponding to the task requests are similar to the resource labels that the nodes have, and no further examples are given here.

In the following, Table 1 and Table 2 are used to illustrate the resource tags that the node has and the resource tags included in the task request.

Assume that the node group includes 6 nodes, which are Node1, Node2, Node3, Node4, Node5, and Node6. According to the resource information of each node, the resource label of each node is determined as shown in Table 1 below. Show.

Table 1 Resource labels of nodes

Based on the resource tags of each node in Table 1, the resource tag set shown in Table 2 below is composed.

Table 2 Resource tag collection

资源类型Resource Type	资源标签Resource tag	资源类型Resource Type	资源标签Resource tag
CPUCPU	CPU主频高High CPU frequency	内存RAM	内存DDR3Memory DDR3
CPUCPU	CPU主频低Low CPU frequency	内存RAM	内存DDR4Memory DDR4
GPUGPU	GPU P40 1块GPU P40	GPUGPU	GPU P100 1块GPU, P100, 1 block
GPUGPU	GPU P40 4块GPU, P40, 4 blocks	GPUGPU	GPU P100 2块GPU P100 2 blocks
GPUGPU	GPU P4GPU P4	带宽bandwidth	带宽高High bandwidth
GPUGPU	GPU V100GPU V100	带宽bandwidth	带宽低Low bandwidth
内存RAM	内存DDR1Memory DDR1	机房engine room	机房1Computer room 1
内存RAM	内存DDR2Memory DDR2	机房engine room	机房2Computer room 2

The client selects the resource tag corresponding to the task request from the resource tag set according to the resource tag set of Table 2 and the resources required to execute the task specified by the task request.

For example, task request 1 includes 3 resource tags, which are CPU high frequency, high bandwidth, and GPU P100. Based on the above table 1, only the 4 resource tags included in Node 3 include high CPU frequency and high bandwidth. , GPU 100, then Node 3 is the target node that can execute the task specified by the task request 1.

For another example, task request 2 includes two resource tags, namely GPU 40 blocks and memory DDR1. Based on the above table 1, the resource tags of Node 2 and Node4 include at least GPU 40 blocks and memory DDR1. The target nodes are determined in 2 and Node4. Specifically, the target nodes can be randomly determined from Node2 and Node4, or the target nodes can be determined based on the resources available in Node2 and Node4, such as selecting a node with more available resources as the target. node.

In this implementation manner, after receiving the task request, the resource scheduling device may determine a target node to execute a task according to whether a node having a resource label including a resource label corresponding to the task request is determined in the node group. The method of matching the resource tags to determine the target node is simple and helps save time.

As another implementation manner, the resources possessed by the node indicated by the resource tag possessed by the target node meet the requirements of the resources indicated by the resource tag corresponding to the task request.

Specifically, when the resource scheduling device receives the task request, it determines from a node group the resources that the node indicated by the resource label has, and the node that meets the resource requirements indicated by the resource label corresponding to the task request, that is, Target node.

In this implementation, a node may have one or more resource tags corresponding to the following resource types: number of CPU cores, memory capacity, disk capacity, and bandwidth. Exemplarily, the two resource tags possessed by node 4 respectively indicate the available disk capacity of 201-300GB and the memory capacity of 1000-1500MB. The resource labels corresponding to the task requests are similar to the resource labels that the nodes have, and no further examples are given here.

In the following, Table 3 and Table 4 are used to illustrate the resource tags that the node has and the resource tags included in the task request.

Assume that the node group includes 6 nodes, which are Node1, Node2, Node3, Node4, Node5, and Node6. Based on the resource information of each node, the resource label of each node is determined as shown in Table 3 below. Show.

Table 3 Resource labels of nodes

Based on the resource tags of each node in Table 3, the resource tag set shown in Table 4 below is composed.

Table 4 resource tag collection

The client selects the resource tag corresponding to the task request from the resource tag set according to the resource tag set of Table 4 and the resources required to execute the task specified by the task request.

In this implementation manner, the resource label that the node has and the resource label corresponding to the task request may be the same or different.

For example, task request 1 includes a resource tag indicating a memory capacity of 500-700 MB. Based on the above Table 3, the resources indicated by the resource tags of Node 2 and Node 5 include the memory capacity indicated by the resource tag corresponding to task request 1. 500-1000MB, so Node 2 and Node 5 can be used as target nodes. In addition, although the resource indicated by the three resource tags of Node 4 does not include the memory capacity indicated by the resource tag corresponding to task request 500-1000MB, the memory capacity indicated by the resource tag of Node 4 has 1000-1500MB, The memory capacity that is higher than the task requires 500-1000MB, that is, Node 4 meets the resource requirements indicated by the resource label corresponding to task request 1, so Node 4 can also be used as the target node. Therefore, a node can be selected from all Nodes 2, 5 and 4 to perform the task specified by the task request 1.

For another example, task request 2 includes two resource tags, which respectively indicate a memory capacity of 500-900MB and a bandwidth of 21-50M. Because only the resources indicated by the two resource tags of Node 2 meet the instructions of the resource tags corresponding to the task request 2 Resource requirements, so this Node 2 can be used as the target node.

In this way, the resources possessed by the node indicated by the resource tag possessed by the node satisfy the requirements of the resource indicated by the resource tag corresponding to the task request, and the target node can be determined without the need of the resource tag and The resource labels corresponding to the task requests are exactly the same to determine the target node, so that the task can be scheduled more flexibly.

Further, in this implementation manner, if the available resources in a node change, the resource labels of the nodes shown in Table 3 also change, so when the available resources of the node change, it is necessary to target Table 3 and the table when the available resources of the node change. The resource label in 4 is updated. Specifically, the resource label of the node may be updated according to the available resources of the node.

In combination with the above two implementation manners, as another implementation manner, the resource tag possessed by the node may be a combination of any one or more of the resource tags in Table 1 and any one or more of the resource tags in Table 3. Correspondingly, resources The tag set may also be a combination of the resource tags in Tables 2 and 4 above. For example, Node 1 has three resource tags, which indicate the CPU's main frequency is high, the number of CPU cores is 32, and the bandwidth is 21-50M.

Further, as a possible implementation manner, the task request records the amount of resources required to execute the task. In a specific manner of implementing step 203 above, the resource scheduling device selects from the node group the resource amount that satisfies the task request specification. Target node.

Specifically, the resource scheduling device selects, from the node group, a node having a resource tag corresponding to the task request and satisfying the amount of resources specified by the task request, that is, a target node. In this way, the determined target node not only meets the resources required to execute the task specified by the resource tag included in the task request, but also meets the amount of resources required to execute the task, thereby helping to improve the execution efficiency of the task specified by the task request.

Step 204: The resource scheduling device instructs the target node to execute a task specified by the task request.

Through the above steps 201 to 204, the client generates a task request. The task request specifies the resources required to perform the task through the resource tag included in the task request. The client sends a task request to the resource scheduling device, and the resource scheduling device receives the task sent by the client. Request, the resource scheduling device determines a target node from the node group according to the resource label that each node in the node group has and the resource label included in the task request; the target node has the resource label included in the task request; resource scheduling The device instructs the target node to execute a task specified by the task request. On the one hand, since the task request includes a resource tag corresponding to the task request, the task request specifies a resource required to execute the task through the resource tag included in the task request, so the resource required by the task specified by the task request To determine the target node, a node with a processing capacity suitable for the task specified by the task request can be found, thereby helping to improve the execution efficiency of the task specified by the task node. On the other hand, when determining the target node to execute the task specified by the task request, the resource label included in the task request is determined from the node group according to the resource label that the node has. In this way, the Resources, which in turn helps improve the resource utilization of nodes.

Based on any of the above embodiments, the task request may further include a task affinity resource tag, and the task affinity resource tag is used to indicate the affinity or anti-affinity of the task specified by the task request and another task, thereby On-demand scheduling of tasks can be further implemented based on the affinity between tasks.

As a possible implementation manner, when the task specified by the task request has affinity with another task, the resource scheduling device instructs the target node to perform another task; when the task specified by the task request has anti-affinity with another task At this time, the resource scheduling device instructs another node in the node group different from the target node to perform another task. In this way, when determining a node that executes a task, the resource scheduling device considers the affinity between the tasks, so that a target node suitable for performing the task can be determined.

As an example, for example, the task specified by the task request is task 1, the other task is task 2, and the target node specified by the resource scheduling device for task 1 is Node1. If task 1 and task 2 have affinity, Node 1 can be used to perform task 2.

As another example, for example, the task specified by the task request is task 1, the other task is task 3, and the target node specified by the resource scheduling device for task 1 is Node1. If task 1 and task 3 have anti-affinity, nodes other than Node 1 can be used to perform task 2 to avoid task 1 and task 2 being executed on the same node, thereby improving the processing efficiency of task 1 and task 2.

Based on any of the above embodiments, a flag bit may be added to the task request, and the flag bit is used to identify whether the resource tag corresponding to the task request is a necessary resource tag. Specifically, if the resource tag identified by the flag bit is a necessary resource tag, the target node must have the resource tag identified by the flag bit; if the resource tag identified by the flag bit is an unnecessary resource tag, the target node may not have the resource tag. The tag identifies the resource label.

As an example, the value of the flag bit is 0, which means that the resource tag identified by the flag bit is an unnecessary resource tag; the value of the flag bit is 1, which means that the resource tag identified by the flag bit is a necessary resource tag. As another example, the value of the flag bit is 1, which indicates that the resource tag identified by the flag bit is an unnecessary resource tag; the value of the flag bit is 0, which indicates that the resource tag identified by the flag bit is a necessary resource tag.

As an example, taking a flag value of 1 to indicate a necessary resource tag and a flag value of 0 to indicate an unnecessary resource tag, for example, it is assumed that the resource tag corresponding to the task request includes GPU P40 and memory DDR, where GPU P40 is Necessary resource tag, memory DDR is an unnecessary resource tag, then the flag bit that can be in the task request can be expressed as 1, 0, see Table 1, select the resource with the task request from the nodes shown in Table 1 At this time, according to the resource tag and flag bit corresponding to the task request, Nodes 2 and 4 with the necessary resource tags GPU P40 can be determined, and Node 2 and Node 4 can be used to perform the task. The task specified in the request.

In this application, when a resource scheduling device receives a task request and determines a target node from a node group, because some tasks require resources to include GPUs during execution, and some tasks require resources to execute without GPUs, therefore, If a task without a GPU requirement is scheduled for execution on a node with a GPU, a task that urgently needs GPU resources cannot be scheduled to the node, thereby causing a waste of GPU resources on the node. Therefore, in order to avoid wasting GPU resources in the node group, based on the above method, several alternative implementation methods are provided below.

As a specific example, when the resource tag corresponding to the task request does not include a GPU resource tag, the resource scheduling apparatus selects a node without a GPU resource tag from the node group, and instructs the selected node to execute the task request specification Task.

Specifically, the resource scheduling device selects a node without a resource tag of the GPU from the node group, and selects a target node having a resource tag corresponding to the task request from the nodes without the resource tag of the GPU, and schedules the target node to execute a task The task specified in the request.

As another specific example, when the resource tag corresponding to the task request includes a resource tag of a GPU, the resource scheduling apparatus selects a node having a resource tag of the GPU from a node group, and instructs the selected node to execute the task specified by the task request. task.

Specifically, the resource scheduling device selects a node having a resource tag of the GPU from the node group, and selects a target node having a resource tag corresponding to the task request from the nodes having the resource tag of the GPU, and schedules the target node to execute the task request. Assigned tasks.

With this implementation method, according to whether a resource tag corresponding to a task request includes a resource tag of a GPU, selecting a target node to execute a task specified by the task request, it is possible to implement GPU resources on demand according to the resources required by the task specified by the task request. Scheduling can improve the resource utilization of the nodes in the node group, especially the utilization of GPU resources, and help avoid the waste of redundant GPU resources in the node caused by the exhaustion of CPU resources in the node.

As a specific example, if the amount of resources required to execute a task does not include GPU resources, selecting a node that does not have the GPU resource from the node group, and instructing the selected node to execute the task request specification Task.

As another specific example, if the amount of resources required to execute a task includes GPU resources, selecting a node with the GPU resources from the node group, and instructing the selected node to execute the task specified by the task request task.

A possible implementation manner of the foregoing resource scheduling method is described in detail below with reference to FIG. 4.

FIG. 4 illustrates a schematic flowchart of another resource scheduling method according to an embodiment of the present application. The method includes the following steps:

Step 401: The resource scheduling device receives a task request.

In step 402, the resource scheduling device determines whether a GPU is required to execute the task specified in the task request; if so, step 403 is performed; if not, step 406 is performed.

In step 403, the resource scheduling device traverses the node label database to find the node with the resource label of the GPU. If so, step 404 is performed; if not, step 409 is performed. The node tag database stores a correspondence between a node identifier and a resource tag.

Step 404: The resource scheduling device determines a target node having a resource label corresponding to the task request from the nodes having a resource label of the GPU.

Step 405: The resource scheduling device schedules the target node to execute the task specified by the task request.

In step 406, the resource scheduling device traverses the node tag database to find nodes without resource tags of the GPU. If so, step 407 is performed; if not, step 409 is performed. The node tag database stores a correspondence between a node identifier and a resource tag.

Step 407: The resource scheduling device determines a target node having a resource label corresponding to the task request from the nodes without the resource label of the GPU.

Step 408: The resource scheduling device schedules the target node to execute the task specified by the task request.

Step 409: The resource scheduling device sends a task scheduling failure message to the client, and abandons scheduling the task specified by the task request.

Through this example, the optimal combination scheduling of GPU resources based on user-defined multiple resource tags and sensing nodes is compared to the solution in the prior art where all nodes are treated equally without GPU resources of sensing nodes. The examples in this application On-demand scheduling of GPU resources for tasks can help improve the resource utilization of each node in the node group, especially the utilization of GPU resources of the node.

It can be understood that, in order to implement corresponding functions, each device in the foregoing embodiments may include a hardware structure and / or a software module corresponding to each function. Those skilled in the art should easily realize that, with reference to the units and algorithm steps of each example described in the embodiments disclosed herein, this application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is performed by hardware or computer software-driven hardware depends on the specific application of the technical solution and design constraints. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

In the case of using an integrated unit, FIG. 5 shows a possible exemplary block diagram of a device involved in the embodiment of the present application, and the device 500 may exist in the form of software. The apparatus 500 may include a communication unit 501 and a processing unit 502. As an implementation manner, the communication unit 501 may include a receiving unit and a sending unit. The processing unit 502 is configured to control and manage the operations of the device 500. The communication unit 501 is configured to support communication between the device 500 and other devices (such as a client). The device 500 may further include a storage unit 503 for storing the program code and data of the device 500, such as the resource tag of the node in the above method.

The processing unit 502 may be a processor or a controller. For example, the processing unit 502 may be a general-purpose CPU, a general-purpose processor, digital signal processing (DSP), application-specific integrated circuits (ASIC), and a field programmable gate. Array (field programmable array, FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logical blocks, modules, and circuits described in connection with the present disclosure. The processor may also be a combination that realizes a computing function, for example, includes a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The communication unit 501 may be a communication interface, a transceiver, or a transceiver circuit. The communication interface is collectively referred to. In a specific implementation, the communication interface may include multiple interfaces. The storage unit 503 may be a memory.

The device 500 may also be a resource scheduling device involved in this application. The processing unit 502 may support the device 500 to perform the actions of the resource scheduling device in the method examples above. For example, the processing unit 502 may perform

steps

203 and 204 in FIG. 2 . The communication unit 501 may support communication between the device 500 and a client. For example, the communication unit 501 is configured to support the device 500 to perform step 202 in FIG. 2.

Specifically, the communication unit 501 may be configured to receive a task request sent by a client, where the task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to perform a task by using the resource tag included in the task request. ;

The processing unit 502 may be configured to determine a target node from the node group according to a resource label of each node in the node group, where the target node has the resource label included in the task request; instruct the target node to execute The task requests a specified task.

In a possible implementation manner, the resource tag includes a resource tag used to identify a GPU; the processing unit 502 is further configured to: when the resource tag corresponding to the task request does not include a resource tag of the GPU And selecting a node from the node group that does not have a resource tag of the GPU; and instructing the selected node to execute a task specified by the task request.

In a possible implementation manner, the task request records the amount of resources required to perform the task; the processing unit 502 is configured to: select from the node group a resource amount that satisfies the task request specification Target node.

In a possible implementation manner, the processing unit 502 is further configured to: obtain resource information of each node in the node group, where the resource information records a node identifier of the node and resources possessed by the node; according to the The resource information of each node in the node group records the resource tag and the amount of resources that each node in the node group has.

In a possible implementation manner, the processing unit 502 is further configured to: when the task specified by the task request has affinity with another task, instruct the target node to execute the another task; in When the task specified by the task request has anti-affinity with another task, another node in the node group different from the target node is instructed to execute the another task.

The apparatus 500 may also be a client involved in this application. The processing unit 502 may support the device 500 to perform the actions of the client in the method examples above. For example, the processing unit 502 is configured to support the device 500 to perform step 201 in FIG. 2. The communication unit 501 may support communication between the device 500 and other devices (such as a resource scheduling device). For example, the communication unit 501 is used to support the device 500 to perform step 202 in FIG. 2.

Specifically, the processing unit 502 may be configured to generate a task request, where the task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to perform a task through the resource tag included in the task request;

The communication unit 501 may be configured to send the task request to a resource scheduling device, where the task request is used by the resource scheduling device to determine a target for executing a task specified by the task request according to a resource label that each node in the node group has. Node, the target node has a resource tag included in the task request.

Refer to FIG. 6, which is a schematic diagram of a device provided in the present application. The device may be the resource scheduling device or a client described above. The device 600 includes: a memory 601, a processor 602, and a communication interface 603. Optionally, the device 600 may further include a bus 604. The communication interface 603, the processor 602, and the memory 601 can be connected to each other through a bus 604. The bus 604 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (referred to as an abbreviation). EISA) bus and so on. The bus 604 can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only a thick line is used in FIG. 6, but it does not mean that there is only one bus or one type of bus.

The processor 602 may be a CPU, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of the program of the solution of the present application.

The communication interface 603 uses any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), and wireless local area networks (WLAN). Wired access network, etc.

The memory 601 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (random access memory, RAM), or other types that can store information and instructions The dynamic storage device can also be electrically erasable and programmable read-only memory (Erasable, Programmable, read-only memory), compact disc (read-only memory, CD-ROM) or other optical disk storage, optical disk storage (including compression Optical discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer Any other medium, but not limited to this. The memory may exist independently, and is connected to the processor through the bus 604. The memory can also be integrated with the processor.

The memory 601 is configured to store a computer execution instruction for executing the solution of the present application, and the processor 602 controls execution. The processor 602 is configured to execute a computer execution instruction stored in the memory 601, so as to implement the method provided by the foregoing embodiment of the present application.

Optionally, the computer-executable instructions in the embodiments of the present application may also be referred to as application program codes, which are not specifically limited in the embodiments of the present application.

The above product can execute the method provided in the embodiment of the present application, and has the corresponding functional modules and beneficial effects of executing the method. For technical details not described in detail in this embodiment, reference may be made to the method provided in the embodiment of the present application.

In the above embodiments, all or part may be implemented by software, hardware, or a combination thereof. When implemented using a software program, it may be all or partly implemented in the form of a computer program product. A computer program product includes one or more instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are wholly or partially generated. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable device. The instructions may be stored on a computer storage medium or transmitted from one computer storage medium to another computer storage medium. For example, the instructions may be wired (e.g., coaxial cable, fiber optic, twisted pair, etc.) from a website site, computer, server, or data center. Wire) or wireless (such as infrared, wireless, microwave, etc.) to another website site, computer, server or data center. The computer storage medium may be any medium that can be accessed by a computer or a data storage device such as a server, a data center, or the like that includes one or more media integrations. The medium may be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape, a magneto-optical disk (MO), etc.), an optical medium (for example, an optical disk), or a semiconductor medium (for example, ROM, EPROM, EEPROM, solid state disk (SSD)) )Wait.

Embodiments of the present application are described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and combinations of processes and / or blocks in the flowcharts and / or block diagrams can be implemented by instructions. These instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine such that instructions executed by the processor of a computer or other programmable data processing device are generated for implementation Means of the function specified in one block or blocks of the flowchart or block and block diagrams.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to work in a particular manner such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, the instructions The device implements the functions specified in one or more flowcharts and / or one or more blocks of the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing device, so that a series of steps can be performed on the computer or other programmable device to produce a computer-implemented process, which can be executed on the computer or other programmable device. The instructions provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks of the block diagrams.

Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. In this way, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these changes and variations.

Claims

A resource scheduling method includes:

Receiving a task request sent by a client, the task request including a resource tag corresponding to the task request, and the task request specifying a resource required to perform a task through the resource tag included in the task request;

Determining a target node from the node group according to the resource label that each node in the node group has, the target node having the resource label included in the task request;

Instruct the target node to execute a task specified by the task request.
The method according to claim 1, further comprising:

When the resource tag corresponding to the task request does not include the resource tag of the GPU, selecting a node without the resource tag of the GPU from the node group;

Instruct the selected node to execute the task specified by the task request.
The method according to claim 1 or 2, wherein the task request records an amount of resources required to perform the task;

The determining a target node from the node group according to a resource label of each node in the node group includes:

Select a target node from the node group that satisfies the amount of resources specified by the task request.
The method according to any one of claims 1 to 3, wherein the method further comprises:

Acquiring resource information of each node in the node group, where the resource information records a node identifier of the node and resources possessed by the node;

According to the resource information of each node in the node group, a resource label and an amount of resources that each node in the node group has are recorded.
The method according to any one of claims 1 to 4, further comprising:

Instructing the target node to execute the another task when the task specified by the task request has affinity with another task;

When the task specified by the task request has anti-affinity with another task, another node in the node group different from the target node is instructed to perform the another task.
A resource scheduling method includes:

Generating a task request, the task request including a resource tag corresponding to the task request, and the task request specifying a resource required to perform a task by using the resource tag included in the task request;

Sending the task request to a resource scheduling device, where the task request is used by the resource scheduling device to determine a target node to perform a task specified by the task request according to a resource label that each node in the node group has, and the target node Having a resource tag included in the task request.
A device, comprising:

A communication unit, configured to receive a task request sent by a client, the task request including a resource tag corresponding to the task request, and the task request specifying a resource required to perform a task by using the resource tag included in the task request;

A processing unit, configured to determine a target node from the node group according to a resource label that each node in the node group has, the target node having the resource label included in the task request; and instructing the target node to perform the task Request the specified task.
The apparatus according to claim 7, wherein the processing unit is further configured to:

When the resource tag corresponding to the task request does not include the resource tag of the GPU, selecting a node without the resource tag of the GPU from the node group;

Instruct the selected node to execute the task specified by the task request.
The apparatus according to claim 7 or 8, wherein the task request records an amount of resources required to perform the task; and the processing unit is specifically configured to:

Select a target node from the node group that satisfies the amount of resources specified by the task request.
The device according to any one of claims 7 to 9, wherein the processing unit is further configured to:

Acquiring resource information of each node in the node group, where the resource information records a node identifier of the node and resources possessed by the node;

According to the resource information of each node in the node group, a resource label and an amount of resources that each node in the node group has are recorded.
The device according to any one of claims 7 to 10, wherein the processing unit is further configured to:

Instructing the target node to execute the another task when the task specified by the task request has affinity with another task;

When the task specified by the task request has anti-affinity with another task, another node in the node group different from the target node is instructed to perform the another task.
A device, comprising:

A processing unit, configured to generate a task request, where the task request includes a resource tag corresponding to the task request, and the task request specifies a resource required to perform a task by using the resource tag included in the task request;

A communication unit, configured to send the task request to a resource scheduling device, where the task request is used by the resource scheduling device to determine a target node to perform a task specified by the task request according to a resource tag possessed by each node in the node group , The target node has a resource tag included in the task request.
A device, comprising a processor and a memory; the processor executes computer instructions stored in the memory, so that the device executes the method according to any one of claims 1 to 6.
A computer-readable storage medium, characterized in that computer instructions are stored in the computer-readable storage medium, and the computer instructions are executed by a computer to implement the method according to any one of claims 1 to 6.