Job execution control program, information processing device and job execution control method

By calculating the deviation in execution times between local and remote memory groups, the job scheduler optimally allocates jobs to minimize waiting and performance loss due to remote memory latency.

JP2024163686AInactive Publication Date: 2024-11-22FUJITSU LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023079512
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-12
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The latency when accessing remote memory is greater than local memory, leading to a decline in job execution performance, especially when the memory requirements of a job exceed the local memory capacity, necessitating the use of remote memory.

Method used

A job scheduler determines the execution node by calculating the deviation between the execution times of jobs using local and remote memory groups, allocating jobs to nodes based on whether the deviation satisfies a predetermined condition to minimize performance degradation.

Benefits of technology

This approach suppresses the occurrence of job waiting and deterioration in execution performance by optimizing the allocation of jobs to nodes with or without remote memory.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024163686000001_ABST
    Figure 2024163686000001_ABST
Patent Text Reader

Abstract

To provide a job execution control program, an information processing device, and a job execution control method capable of suppressing a decrease in execution performance of each job.SOLUTION: By referring to a storage unit that stores first execution time for each job using a first memory group and second execution time for each job using a second memory group that takes longer time to access than the first memory group, first execution time and second execution time corresponding to a specific job are specified. A value indicating a discrepancy degree between the specified first execution time and the second execution time is calculated. A specific memory group for executing the specific job is determined from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition.SELECTED DRAWING: Figure 18
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a job execution control program, an information processing apparatus, and a job execution control method. [Background technology]

[0002] A business providing a service to a user (hereinafter, simply referred to as a business) constructs, for example, a business system required for providing the service. In such a business system, the business operates, for example, a job scheduler that determines the execution time of each job.

[0003] Specifically, for example, when a new job is submitted, the job scheduler identifies nodes capable of executing the new job based on information such as the execution status of the job at each node. Then, the job scheduler issues an instruction to execute the new job to one of the identified nodes (for example, see Patent Document 1). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 8-161225 Summary of the Invention [Problem to be solved by the invention]

[0005] Here, if the amount of memory required to execute each job is greater than the amount of memory (hereinafter also referred to as local memory) installed in each node (hereinafter also referred to as simply execution node), the job scheduler, for example, allocates external memory (hereinafter also referred to as remote memory) such as CXL (Compute Express Link) memory to one of the nodes. Then, the node to which the external memory is allocated executes each job by using the allocated remote memory in addition to the local memory installed in the node, for example.

[0006] Here, the latency when accessing the remote memory is greater than the latency when accessing the local memory, for example. Therefore, it is preferable that the job scheduler selects either a node to which the remote memory is assigned or a node to which the remote memory is not assigned, for example, according to the execution characteristics of each job. In other words, it is preferable that the job scheduler determines the execution node of each job so that the degradation of the execution performance of each job can be suppressed.

[0007] SUMMARY OF THE DISCLOSURE In one aspect, the present invention provides a job execution control program, an information processing device, and a job execution control method that are capable of suppressing a decrease in the execution performance of each job. [Means for solving the problem]

[0008] One aspect of the embodiment refers to a memory unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that takes longer to access than the first memory group is used, identifies the first execution time and the second execution time corresponding to a specific job, calculates a value indicating the degree of deviation between the identified first execution time and the second execution time, and determines a specific memory group from the first memory group and the second memory group to execute the specific job based on whether the calculated value satisfies a predetermined condition. Effect of the Invention

[0009] According to one aspect, it is possible to suppress a decrease in the execution performance of each job. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating the configuration of an information processing system 10. As shown in FIG. [Diagram 2] FIG. 2 is a diagram for explaining a specific example of the job management information 131 when the CXL memory RM is not used. [Diagram 3] FIG. 3 is a diagram for explaining a specific example of the job management information 131 when the CXL memory RM is used. [Figure 4] FIG. 4 is a diagram illustrating the configuration of the information processing system 10. As shown in FIG. [Diagram 5] FIG. 5 is a diagram illustrating the configuration of the information processing system 10. As shown in FIG. [Figure 6] FIG. 6 is a diagram illustrating the hardware configuration of the information processing device 1. As shown in FIG. [Figure 7] FIG. 7 is a diagram illustrating functions of the information processing device 1 in the first embodiment. [Figure 8] FIG. 8 is a diagram illustrating information stored in the information storage area 130 in the first embodiment. [Figure 9] FIG. 9 is a flow chart outlining the job control process in the first embodiment. [Figure 10] FIG. 10 is a sequence chart illustrating details of the job control process in the first embodiment. [Figure 11] FIG. 11 is a sequence chart illustrating details of the job control process in the first embodiment. [Figure 12] FIG. 12 is a sequence chart illustrating details of the job control process in the first embodiment. [Figure 13]FIG. 13 is a sequence chart illustrating details of the job control process in the first embodiment. [Figure 14] FIG. 14 is a sequence chart illustrating details of the job control process in the first embodiment. [Figure 15] FIG. 15 is a sequence chart illustrating details of the job control process in the first embodiment. [Figure 16] FIG. 16 is a diagram illustrating a specific example of the execution count information 133. As shown in FIG. [Figure 17] FIG. 17 is a diagram illustrating a specific example of the execution count information 133. As shown in FIG. [Figure 18] FIG. 18 is a diagram for explaining a specific example of the execution time information 132. As shown in FIG. [Figure 19] FIG. 19 is a diagram illustrating a specific example of the execution count information 133. As shown in FIG. [Figure 20] FIG. 20 is a diagram for explaining a specific example of the execution time information 132. As shown in FIG. [Figure 21] FIG. 21 is a diagram for explaining a specific example of the execution time information 132. As shown in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. However, such descriptions should not be interpreted in a limiting sense and do not limit the subject matter described in the claims. In addition, various changes, substitutions, and modifications can be made without departing from the spirit and scope of the present disclosure. In addition, different embodiments can be appropriately combined.

[0012] [Configuration of the information processing system in the first embodiment] First, a description will be given of the configuration of the information processing system 10. Fig. 1, Fig. 4, and Fig. 5 are diagrams for explaining the configuration of the information processing system 10.

[0013] The information processing system 10 includes, for example, an information processing device 1, a node 2a, a node 2b, a node 2c, a node 2d, a switch 3, a memory housing 4a, a memory housing 4b, a memory housing 4c, and a memory housing 4d. Hereinafter, the nodes 2a, 2b, 2c, and 2d are collectively referred to simply as nodes 2. Also, the memory housings 4a, 4b, 4c, and 4d are collectively referred to simply as memory housings 4. Hereinafter, a case in which the information processing system 10 includes four nodes 2 will be described, but the information processing system 10 may include, for example, a number of nodes 2 other than four. Also, a case in which the information processing system 10 includes four memory housings 4 will be described, but the information processing system 10 may include, for example, a number of memory housings 4 other than four.

[0014] The information processing device 1 is, for example, a physical machine, and is an information processing device 1 on which a program 1a (hereinafter also referred to as a job scheduler 1a) that schedules the execution timing of jobs executed in each node 2 runs.

[0015] The node 2 is, for example, a physical machine, and executes jobs according to the scheduling performed by the job scheduler 1a. Note that the node 2 may be, for example, a node constituting an HPC (High Performance Computing) system.

[0016] Specifically, in the example shown in FIG. 1, node 2a is equipped with two local memories LM, node 2b is equipped with two local memories LM, node 2c is equipped with four local memories LM, and node 2d is equipped with four local memories LM.

[0017] The switch 3 is, for example, a device that connects each node 2 to each memory chassis 4. In the following, the description will be given assuming that the switch 3 is a CXL switch 3.

[0018] The memory case 4 is, for example, a case that stores a remote memory RM that can be accessed from each node 2. In the following, the remote memory RM will be described as being a CXL memory RM.

[0019] Specifically, in the example shown in FIG. 1, memory housing 4a stores three CXL memories RM, memory housing 4b stores three CXL memories RM, memory housing 4c stores two CXL memories RM, and memory housing 4d stores two CXL memories RM.

[0020] Then, for example, when starting execution of a job to be executed (hereinafter also referred to as a job to be executed or a specific job), if the amount of memory required to execute the job to be executed is greater than the amount of memory (local memory LM) installed in the execution node 2 of the job to be executed, the job scheduler 1a allocates the necessary CXL memory RM to the execution node 2 of the job to be executed.

[0021] This enables the job scheduler 1a to execute the job to be executed even if, for example, the amount of memory required to execute the job to be executed is greater than the amount of memory installed in the execution node 2 of the job to be executed.

[0022] [Specific example of execution of a job when not using CXL memory] Next, a specific example of execution of a job to be executed without using the CXL memory RM will be described. Specifically, a specific example of execution of each job to be executed by using the local memory LM mounted on each node 2 will be described.

[0023] [Specific example of job management information when not using CXL memory RM] First, a specific example of job management information 131 when the CXL memory RM is not used will be described. Fig. 2 is a diagram for explaining a specific example of job management information 131 when the CXL memory RM is not used. The job management information 131 is, for example, information generated by the job scheduler 1a, and is information including the memory amount and execution node 2 required for the execution of each job to be executed.

[0024] In the following description, it is assumed that each local memory LM and each CXL memory RM is a memory having a storage area of ​​32 GB. In the following description, it is assumed that the nodes 2a and 2b are nodes 2 to which the CXL memory RM may be allocated, and the nodes 2c and 2d are nodes 2 to which the CXL memory RM is not allocated.

[0025] 2 has, for example, items such as "job ID," in which a job ID identifying each job to be executed is set, "memory amount," in which the amount of memory required to execute each job to be executed (for example, the amount of memory specified in advance by the worker who executes each job to be executed) is set, "execution node," in which the execution node 2 of each job to be executed is set, and "execution state," in which the execution state of each job to be executed is set. For example, the "execution state" is set to "executing," which indicates that each job to be executed is being executed, or "waiting," which indicates that each job to be executed is waiting (awaiting execution).

[0026] Specifically, in the job management information 131 shown in FIG. 2, for example, the information in the first line has the "job ID" set to "10," the "memory amount" set to "128 GB," the "execution node" set to "node 2c," and the "status" set to "running."

[0027] That is, in the example shown in FIG. 1, node 2c is equipped with four local memories LM (128 GB local memories LM). Moreover, in the job management information 131 shown in FIG. 2, "128 GB" is set as the "memory amount" of the job to be executed whose "job ID" is "10". Therefore, for example, when the local memory LM equipped in node 2c is not being used for other purposes, the job scheduler 1a determines that the job to be executed whose "job ID" is "10" can be executed in node 2c, and causes node 2c to execute the job to be executed whose "job ID" is "10".

[0028] In addition, in the job management information 131 shown in FIG. 2, for example, in the information on the second line, the "job ID" is set to "11", the "memory amount" is set to "128 GB", the "execution node" is set to "node 2d", and the "status" is set to "running".

[0029] That is, in the example shown in FIG. 1, the node 2d is equipped with four local memories LM (128 GB local memories LM). Moreover, in the job management information 131 shown in FIG. 2, "128 GB" is set as the "memory amount" of the job to be executed whose "job ID" is "11". Therefore, for example, when the local memory LM equipped in the node 2d is not being used for other purposes, the job scheduler 1a determines that the job to be executed whose "job ID" is "11" can be executed in the node 2d, and causes the node 2d to execute the job to be executed whose "job ID" is "11".

[0030] In addition, in the job management information 131 shown in FIG. 2, for example, in the information on the third line, the "job ID" is set to "12", the "memory amount" is set to "128 GB", the "execution node" is set to "node 2c", and the "status" is set to "waiting".

[0031] That is, in the example shown in Fig. 1, node 2c is equipped with four local memories LM (128 GB local memories LM). Also, in the job management information 131 shown in Fig. 2, "128 GB" is set as the "memory amount" of the job to be executed whose "job ID" is "10". Therefore, for example, when the local memory LM equipped in node 2c is not being used for other purposes, the job scheduler 1a determines that the job to be executed whose "job ID" is "10" can be executed in node 2c.

[0032] However, the job management information 131 shown in FIG. 2 indicates, for example, that all of the local memory LM installed in node 2c is being used to execute another job to be executed (a job with a "job ID" of "10"). The job management information 131 shown in FIG. 2 also indicates, for example, that all of the local memory LM installed in node 2d is being used to execute another job to be executed (a job with a "job ID" of "11"). The job management information 131 shown in FIG. 2 further indicates, for example, that the amount of local memory LM installed in each of nodes 2a and 2b is small, and that the job to be executed with a "job ID" of "12" cannot be executed in node 2a or node 2b.

[0033] Therefore, the job scheduler 1a, for example, causes the execution of a job to be executed, whose "job ID" is "12", to wait until the execution of another job to be executed (whose "job ID" is "10") currently being executed in node 2c has finished.

[0034] [Specific example of execution of a job to be executed when using CXL memory] Next, a specific example of execution of a job to be executed when using the CXL memory RM will be described. Specifically, a specific example of execution of each job to be executed using the local memory LM and the CXL memory RM mounted on each node 2 will be described.

[0035] [Specific example of job management information when using CXL memory RM] First, a specific example of job management information 131 when using CXL memory RM will be described. Fig. 3 is a diagram for explaining a specific example of job management information 131 when using CXL memory RM. Below, differences from job management information 131 explained in Fig. 2 will be explained.

[0036] In the job management information 131 shown in FIG. 3, for example, in the information on the third line, the "job ID" is set to "12", the "memory amount" is set to "128 GB", the "execution node" is set to "node 2a", and the "status" is set to "running".

[0037] That is, the job scheduler 1a assigns, for example, one CXL memory RM stored in the memory casing 4a (the CXL memory RM corresponding to the dashed line portion in the memory casing 4a) and one CXL memory RM stored in the memory casing 4b (the CXL memory RM corresponding to the dashed line portion in the memory casing 4b) to the node 2a as shown in Fig. 4. Then, the job scheduler 1a causes the node 2a to execute a job to be executed, for example, with a "job ID" of "12".

[0038] As a result, unlike the case described in Fig. 2, the job scheduler 1a can execute, for example, a job to be executed whose "job ID" is "12" in a node 2 (node ​​2a) whose installed local memory LM (64 GB) is less than the memory amount (128 GB) required to execute the job to be executed. As a result, the job scheduler 1a can, for example, suppress the occurrence of waiting for execution of each job to be executed, and can shorten the time required for each job to be executed to finish. Note that hereinafter, the one or more local memories LM installed in each node 2 and the one or more CXL memories RM allocated to each node 2 are collectively referred to simply as a memory group.

[0039] 5, the latency when the CPU (Central Processing Unit) 21 of a node 2 accesses the CXL memory RM stored in the memory case 4 is greater than, for example, the latency when the CPU 21 of the node 2 accesses the local memory LM mounted on the same node 2. Therefore, even if there are multiple nodes 2 that can use the same memory amount, the execution time of a job to be executed in a node 2 to which a CXL memory RM is assigned is longer than, for example, the execution time of a job to be executed in a node 2 to which a CXL memory RM is not assigned. Therefore, it is preferable that the job scheduler 1a executes many jobs to be executed in, for example, a node 2 to which a CXL memory RM is not assigned.

[0040] However, when many jobs to be executed are executed in a node 2 to which the CXL memory RM is not allocated, for example, many jobs to be executed are generated in the node 2 to which the CXL memory RM is not allocated, and it is possible that each job to be executed cannot be executed efficiently. Therefore, it is preferable that the job scheduler 1a executes some of the jobs to be executed in a node 2 to which the CXL memory RM may be allocated, within a range in which it is possible to suppress a decrease in the execution performance of each job to be executed. Therefore, it is preferable that the job scheduler 1a determines the execution node 2 of each job to be executed to be either a node 2 to which the CXL memory RM may be allocated (for example, node 2a or node 2b in FIG. 4) or a node 2 to which the CXL memory RM is not allocated (for example, node 2c or node 2d in FIG. 4), depending on, for example, the execution characteristics of each job to be executed.

[0041] Therefore, the job scheduler 1a in this embodiment refers to a storage unit that stores the execution time of each job to be executed when the first memory group is used (hereinafter, also referred to as the first execution time) and the execution time of each job to be executed when the second memory group that requires more time to access than the first memory group is used (hereinafter, also referred to as the second execution time), and identifies the first execution time and the second execution time corresponding to the job to be executed. The first memory group is, for example, a memory group corresponding to a node 2 to which the CXL memory RM is not assigned (for example, a memory group consisting of one or more local memories LM). The second memory group is, for example, a memory group corresponding to a node 2 to which the CXL memory RM may be assigned (for example, a memory group consisting of one or more local memories LM, or a memory group consisting of one or more local memories LM and one or more CXL memories RM).

[0042] Then, the job scheduler 1a calculates, for example, a value indicating the degree of deviation between the identified first execution time and the second execution time (hereinafter, simply referred to as a value). Specifically, the job scheduler 1a calculates, for example, the ratio of the first execution time to the second execution time as the value.

[0043] Thereafter, the job scheduler 1a determines a memory group (hereinafter also referred to as a specific memory group) for executing the target job from the first memory group and the second memory group, for example, based on whether the calculated value satisfies a predetermined condition. Specifically, the job scheduler 1a determines to execute the target job by using the first memory group when the value indicating the deviation degree is less than a threshold value (hereinafter also referred to as a first threshold value), and determines to execute the target job by using the second memory group when the value indicating the deviation degree is equal to or greater than the first threshold value.

[0044] That is, the job scheduler 1a in this embodiment judges the magnitude of the deviation between the first execution time and the second execution time corresponding to each job, for example, by referring to the first execution time and the second execution time stored in advance in the storage unit. Then, the job scheduler 1a estimates, for example, that an execution target job for which it has been determined that the deviation is large is an execution target job whose execution performance will be significantly degraded if executed in the node 2 to which the CXL memory RM is assigned (in other words, an execution target job that is not suitable for execution in the node 2 to which the CXL memory RM is assigned). Also, the job scheduler 1a estimates, for example, that an execution target job for which it has been determined that the deviation is not large is an execution target job whose execution performance will not be significantly degraded even if executed in the node 2 to which the CXL memory RM is assigned (in other words, an execution target job that is suitable for execution in the node 2 to which the CXL memory RM is assigned).

[0045] Then, the job scheduler 1a decides to execute a job to be executed that is determined to be unsuitable for execution in a node 2 to which the CXL memory RM is allocated in a node 2 to which the CXL memory RM is not allocated (for example, node 2c or node 2d in FIG. 4). Also, the job scheduler 1a decides to execute a job to be executed that is determined to be suitable for execution in a node 2 to which the CXL memory RM is allocated in a node 2 to which the CXL memory RM may be allocated (for example, node 2a or node 2b in FIG. 4).

[0046] As a result, the job scheduler 1a in this embodiment can, for example, suppress the occurrence of waiting for execution of jobs to be executed in each node 2, and can also suppress a decrease in the execution performance of each job to be executed.

[0047] [Hardware configuration of information processing system 10] Next, a description will be given of the hardware configuration of the information processing system 10. FIG 6 is a diagram illustrating the hardware configuration of the information processing device 1.

[0048] 6, the information processing device 1 includes, for example, a CPU (Central Processing Unit) 101, which is a processor, a memory 102, a communication device (I / O interface) 103, and a storage device 104. Each unit is connected to each other via a bus 105.

[0049] The storage device 104 has, for example, a program storage area (not shown) that stores a program 110 for performing a process for controlling the execution of a job to be executed (hereinafter also referred to as a job control process). The storage device 104 also has, for example, a storage unit 130 (hereinafter also referred to as an information storage area 130) that stores information used when performing the job control process. The storage device 104 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD).

[0050] The CPU 101 executes, for example, a program 110 loaded from the storage device 104 to the memory 102 to perform job control processing.

[0051] The communication device 103 communicates with each node 2 via a network (not shown) such as the Internet.

[0052] [Functions of the communication system in the first embodiment] Next, functions of the information processing system 10 in the first embodiment will be described. Fig. 7 is a diagram for explaining functions of the information processing device 1 in the first embodiment. Fig. 8 is a diagram for explaining information stored in the information storage area 130 in the first embodiment.

[0053] As shown in FIG. 7, the information processing device 1 realizes various functions, including a job receiving unit 111, a time determination unit 112, a deviation calculation unit 113, a memory group determination unit 114, a job execution unit 115, an information management unit 116, and a memory group redetermining unit 117, as functions of a job scheduler 1a, by organic cooperation between hardware such as a CPU 101 and a memory 102 and a program 110.

[0054] 8, the information storage area 130 stores, for example, job management information 131, execution time information 132, and execution count information 133. The execution count information 133 will be described later.

[0055] The job receiving unit 111 receives, for example, an instruction to execute a job to be executed, which is given by an operator who executes each job to be executed via an operation terminal (not shown).

[0056] The time determination unit 112, for example, refers to the execution time information 132 stored in the information storage area 130, and determines the first execution time and the second execution time corresponding to the execution target job for which the job receiving unit 111 has received an execution instruction. The execution time information 132 is information including, for example, the first execution time and the second execution time corresponding to each execution target job.

[0057] The deviation calculation unit 113 calculates, for example, a value indicating the degree of deviation between the first execution time and the second execution time identified by the time identification unit 112. Specifically, the deviation calculation unit 113 calculates, for example, the ratio of the first execution time to the second execution time.

[0058] The memory group determination unit 114 determines the execution node 2 for the job to be executed, for which the job receiving unit 111 has accepted an execution instruction, from between a node 2 to which the CXL memory RM is not assigned (a node 2 having a first memory group) and a node 2 to which the CXL memory RM may be assigned (a node 2 having a second memory group), for example, based on whether the value calculated by the deviation calculation unit 113 satisfies a predetermined condition.

[0059] Specifically, for example, if the value calculated by the deviation calculation unit 113 is less than a first threshold, the memory group determination unit 114 determines to execute the job to be executed in a node 2 to which CXL memory RM is not assigned, and if the value calculated by the deviation calculation unit 113 is equal to or greater than the first threshold, the memory group determination unit 114 determines to execute the job to be executed in a node 2 to which CXL memory RM may be assigned.

[0060] The memory group determination unit 114 may determine, for example, the execution node 2 of the job to be executed, and may also determine the time at which the job to be executed (hereinafter, also referred to as the execution start time).

[0061] The job executing unit 115 executes, for example, the job to be executed, for which the job receiving unit 111 has received an execution instruction, in the node 2 determined by the memory group determining unit 114.

[0062] The information management unit 116 stores, for example, the execution node 2 of the job to be executed determined by the memory group determination unit 114 in the information storage area 130 as a part of the job management information 131.

[0063] Specifically, the information management unit 116 stores, in the information storage area 130, as part of the job management information 131, for example, the execution node 2 of the job to be executed, the job ID of the job to be executed, and the memory amount used by the job to be executed.

[0064] The information management section 116 may store the execution start time determined by the memory group determination section 114 as a part of the job management information 131 in the information storage area 130, for example.

[0065] Furthermore, the information management unit 116 stores, for example, the execution time (first execution time or second execution time) of the job to be executed by the job executing unit 115 in the information storage area 130 as a part of the execution time information 132.

[0066] The memory group re-determination unit 117 changes the execution node 2 (the execution node 2 of the execution target job for which the job receiving unit 111 has accepted an execution instruction) determined by the memory group determination unit 114, for example, in accordance with the memory access frequency (the number of memory accesses per unit time) of the execution target job (hereinafter also referred to as a waiting job) waiting to be executed in a node 2 to which a CXL memory RM is not allocated.

[0067] [Outline of job control process in the first embodiment] Next, an overview of the first embodiment will be described with reference to Fig. 9, which is a flow chart for explaining an overview of job control processing in the first embodiment.

[0068] As shown in FIG. 9, for example, the job scheduler 1a refers to the execution time information 132 stored in the information storage area 130 and identifies the first execution time corresponding to the job to be executed (S1).

[0069] Furthermore, the job scheduler 1a, for example, refers to the execution time information 132 stored in the information storage area 130 and identifies the second execution time corresponding to the job to be executed (S2).

[0070] Then, the job scheduler 1a calculates, for example, a value indicating the degree of deviation between the first execution time specified in the process of S1 and the second execution time specified in the process of S2 (S3).

[0071] Thereafter, the job scheduler 1a determines a specific memory group for executing the job to be executed from the first memory group and the second memory group, based on whether the value calculated in the process of S3 satisfies a predetermined condition (S4).

[0072] As a result, the job scheduler 1a in this embodiment can, for example, suppress the occurrence of waiting for execution of jobs to be executed in each node 2, and can also suppress a decrease in the execution performance of each job to be executed.

[0073] [Details of Job Control Processing in the First Embodiment] Next, details of the job control process in the first embodiment will be described. Figures 10 to 15 are sequence charts for explaining the details of the job control process in the first embodiment. Figures 16 to 21 are diagrams for explaining the details of the job control process in the first embodiment.

[0074] [Memory determination process] First, a specific example of a process for determining a memory group (node ​​2) for executing each job to be executed (hereinafter, also referred to as a memory determination process) among the job execution control processes will be described. Fig. 10 and Fig. 11 are sequence charts for explaining the memory determination process.

[0075] As shown in FIG. 10, the job receiving unit 111 waits until it receives an instruction to execute the job to be executed from a worker who is to execute the job to be executed (NO in S11).

[0076] Then, when an execution instruction for a job to be executed is received (YES in S11), the memory group determination unit 114, for example, refers to the execution count information 133 stored in the information storage area 130 and determines whether or not information about the job to be executed corresponding to the received execution instruction is included (S12). The execution count information 133 is information including, for example, the number of times each job to be executed has been executed.

[0077] That is, the memory group determination unit 114 determines, for example, whether or not the execution target job corresponding to the execution instruction received in the process of S11 is an execution target job that has been executed in the past in each node 2. A specific example of the execution count information 133 will be described below.

[0078] [Example of execution count information] 16, 17, and 19 are diagrams for explaining specific examples of the execution count information 133. FIG.

[0079] The job execution count information 133 shown in FIG. 16 etc. has, for example, items such as "execution order," which sets the order in which each job to be executed was first executed, "program ID," which sets the program ID of each program constituting each job to be executed (each program executed in conjunction with the execution of each job to be executed), and "number of executions," which sets the number of times each job to be executed is executed.

[0080] Specifically, in the execution count information 133 shown in FIG. 16, for example, in the information in the first line, the "execution order" is set to "1," the "program ID" is set to "PG1" and "PG2," and the "number of executions" is set to "4 (times)."

[0081] 16, for example, in the information on the second line, "2" is set as the "execution order," "PG1" and "PG5" are set as the "program ID," and "2 (times)" is set as the "number of executions." Explanation of other information included in FIG. 16 will be omitted.

[0082] 16 includes information (information in the second row) about the job to be executed including programs whose program IDs are PG1 and PG5, for example. Therefore, for example, when the program IDs of the programs included in the job to be executed corresponding to the execution instruction received in the process of S11 are PG1 and PG5, the memory group determination unit 114 determines that the execution count information 133 includes information about the job to be executed corresponding to the execution instruction received in the process of S11.

[0083] In this case, the information management unit 116 adds 1 to the value set in the “number of executions” in the information (information on the second line) about the job including each program whose program ID is PG1 and PG5, as shown in the underlined portion of FIG. 17, for example.

[0084] Returning to FIG. 10, if it is determined that information about the job to be executed corresponding to the execution instruction received in the processing of S11 is included in the execution count information 133 (YES in S12), the time determination unit 112, for example, refers to the execution time information 132 stored in the information storage area 130, and determines whether or not each of the execution time (first execution time) when the job to be executed corresponding to the execution instruction received in the processing of S11 is executed in a node 2 (node ​​2 having a first memory group) to which the CXL memory RM is not assigned, and the execution time (second execution time) when the job to be executed corresponding to the execution instruction received in the processing of S11 is executed in a node 2 (node ​​2 having a second memory group) to which the CXL memory RM may be assigned is stored (S14).

[0085] As a result, if it is determined that both the execution time when executed in a node 2 to which CXL memory RM is not allocated and the execution time when executed in a node 2 to which CXL memory RM may be allocated are stored in the execution time information 132 (YES in S14), the time determination unit 112 determines, for example, the execution time when the job to be executed corresponding to the execution instruction received in the processing of S11 is executed in a node 2 to which CXL memory RM is not allocated (S21), as shown in FIG. 11.

[0086] In this case, the time specifying unit 112 specifies the execution time when the job to be executed corresponding to the execution instruction received in the process of S11 is executed in the node 2 to which the CXL memory RM may be allocated (S22). A specific example of the execution time information 132 will be described below.

[0087] [Example of execution time information] 18, 20, and 21 are diagrams for explaining specific examples of the execution time information 132. FIG.

[0088] 18 and the like has, for example, items of "execution order," which sets the order of when each execution target job was first executed, and "program ID," which sets the program ID of each program constituting each execution target job (each program executed with the execution of each execution target job). Also, the execution time information 132 shown in Fig. 18 and the like has, for example, items of "first execution time," which sets the execution time (first execution time) when each execution target job is executed in a node 2 to which a CXL memory RM is not allocated, "second execution time," which sets the execution time (second execution time) when each execution target job is executed in a node 2 to which a CXL memory RM may be allocated, and "access frequency," which sets the memory access frequency when each execution target job is executed. In the "access frequency", for example, the memory access frequency after a predetermined time (for example, T seconds) from the execution start time of each execution target job (hereinafter also referred to as the first memory access frequency), the memory access frequency in the middle between the execution start time and the execution end time of each execution target job (hereinafter also referred to as the second memory access frequency), or the memory access frequency before a predetermined time (for example, T seconds) before the execution end time of each execution target job (hereinafter also referred to as the third memory access frequency). In addition, in the "access frequency", for example, the average value of the first memory access frequency, the second memory access frequency, and the third memory access frequency may be set. In addition, in the "access frequency", for example, all of the first memory access frequency, the second memory access frequency, and the third memory access frequency may be set. In the following description, it is assumed that "-" is set for items for which no information is set.

[0089] Specifically, in the execution time information 132 shown in FIG. 18, for example, in the information in the first line, "1" is set as the "execution order," "PG1" and "PG2" are set as the "program ID," "-" is set as the "first execution time," "-" is set as the "second execution time," and "-" is set as the "access frequency."

[0090] Also, in the execution time information 132 shown in FIG. 18, for example, in the information on the second line, the "execution order" is set to "2", the "program ID" is set to "PG1" and "PG5", the "first execution time" is set to "10s", the "second execution time" is set to "20s", and the "access frequency" is set to "100MB".

[0091] 18, for example, in the information on the fourth line, "4" is set as the "execution order," "PG4" is set as the "program ID," "-" is set as the "first execution time," "20s" is set as the "second execution time," and "200MB" is set as the "access frequency." Explanation of other information included in FIG. 18 will be omitted.

[0092] 18, for example, the "first execution time" and "second execution time" in the information (information in the second line) about the job to be executed including programs with program IDs PG1 and PG5 are set to "10s" and "20s," respectively. Therefore, for example, when the program IDs of the programs included in the job to be executed corresponding to the execution instruction received in the process of S11 are PG1 and PG5, the time identification unit 112 identifies 10 ms as the first execution time in the process of S21 and identifies 20 ms as the second execution time in the process of S22.

[0093] Returning to FIG. 11, the deviation calculation unit 113 calculates, for example, a value indicating the degree of deviation between the execution time identified in the process of S21 and the execution time identified in the process of S22 (S23).

[0094] Specifically, for example, if 10 ms is specified as the first execution time in the processing of S21 and 20 ms is specified as the second execution time in the processing of S22, the deviation calculation unit 113 calculates 0.5, which is the ratio of 10 ms specified in the processing of S21 to 20 ms specified in the processing of S22.

[0095] Thereafter, the memory group determination unit 114 determines whether or not the value calculated in the process of S23 is less than a threshold value (first threshold value) (S24).

[0096] As a result, if it is determined that the value calculated in the processing of S23 is less than the threshold value (YES in S24), the memory group determination unit 114 determines, for example, the execution node 2 of the job to be executed for which an execution instruction was received in the processing of S11 to be a node 2 to which CXL memory RM is not assigned (a node 2 having the first memory group) (S25).

[0097] That is, an execution target job for which the value calculated in the process of S23 is less than the threshold value can be determined to be an execution target job for which, for example, when executed in a node 2 to which CXL memory RM may be allocated, the execution time is likely to be significantly longer than when executed in a node 2 to which CXL memory RM is not allocated. Therefore, the memory group determination unit 114 determines, for example, the execution node 2 for an execution target job for which the value calculated in the process of S23 is less than the threshold value to be a node 2 to which CXL memory RM is not allocated.

[0098] On the other hand, if it is determined that the value calculated in the processing of S23 is greater than or equal to the threshold value (NO in S24), the memory group determination unit 114 determines, for example, the execution node 2 of the job to be executed for which an execution instruction was received in the processing of S11 to be the node 2 to which the CXL memory RM may be assigned (node ​​2 having the second memory group) (S13).

[0099] That is, an execution target job for which the value calculated in the process of S23 is equal to or greater than the threshold value can be determined to be an execution target job for which the execution time does not differ significantly even when executed in a node 2 to which the CXL memory RM is likely to be allocated, as compared to when executed in a node 2 to which the CXL memory RM is not allocated. Therefore, the memory group determination unit 114 determines, for example, the execution node 2 for the execution target job for which the value calculated in the process of S23 is equal to or greater than the threshold value as the node 2 to which the CXL memory RM is likely to be allocated.

[0100] This enables the job scheduler 1a to suppress a decrease in execution performance during the execution of each job to be executed, while suppressing the number of jobs to be executed (i.e., the number of waiting jobs) to be executed in a node 2 to which CXL memory RM is not allocated, for example.

[0101] In addition, the memory group determination unit 114 may determine multiple nodes 2 as execution nodes 2, for example, in the processing of S13 and the processing of S25, depending on the amount of memory required to execute the job to be executed for which an execution instruction was accepted in the processing of S11.

[0102] On the other hand, if it is determined that at least one of the execution time when executed in a node 2 to which the CXL memory RM is not allocated and the execution time when executed in a node 2 to which the CXL memory RM may be allocated is not stored (NO in S14), the time determination unit 112, etc., does not perform, for example, processes S21 to S24.

[0103] Then, the memory group determination unit 114 determines, for example, the execution node 2 of the job to be executed corresponding to the execution instruction received in the process of S11 as the node 2 to which the CXL memory RM may be allocated (S13).

[0104] Thereafter, the information management unit 116 stores, for example, information corresponding to the job to be executed corresponding to the execution instruction received in the processing of S11 in the information storage area 130 as part of the execution count information 133, as shown in the underlined portion of FIG.

[0105] In addition, the memory group determination unit 114 also performs processing of S13 when, for example, it is determined that information about the job to be executed corresponding to the execution instruction received in processing of S11 is not included in the execution count information 133 (NO in S12).

[0106] Then, for example, after the process of S13 or the process of S25, the memory group determination unit 114 ends the memory determination process.

[0107] [Job execution processing] Next, a specific example of the process of executing each job to be executed (hereinafter, also referred to as a job execution process) among the job execution control processes will be described. Fig. 12 and Fig. 13 are sequence charts for explaining the job execution process.

[0108] 12, the job execution unit 115 waits until the job execution timing (NO in S101), for example. The job execution timing may be the execution start time of each job to be executed (the execution start time included in the job management information 131), for example.

[0109] Then, when the job execution timing comes (YES in S101), the job execution section 115 executes the execution target job that has become the job execution timing in the process of S101 (S102).

[0110] Specifically, the job execution unit 115, for example, refers to the job management information 131 stored in the information storage area 130, and identifies the execution node 2 of the job to be executed that has reached its job execution timing in the process of S101. Then, the job execution unit 115 causes the identified execution node 2 to execute the job to be executed that has reached its job execution timing in the process of S101.

[0111] Next, the job execution unit 115, for example, refers to the execution time information 132 stored in the information storage area 130, and determines whether or not the execution time (first execution time) of the job to be executed in the processing of S102 when it is executed in a node 2 (node ​​2 having a first memory group) to which the CXL memory RM is not assigned, and the execution time (second execution time) of the job to be executed in the processing of S102 when it is executed in a node 2 (node ​​2 having a second memory group) to which the CXL memory RM may be assigned are stored (S103).

[0112] As a result, if it is determined that at least one of the execution time when the job to be executed in the processing of S102 is executed in a node 2 to which CXL memory RM is not allocated and the execution time when the job to be executed in the processing of S102 is executed in a node 2 to which CXL memory RM may be allocated is not stored (NO in S103), the job execution unit 115 executes, for example, the job to be executed that has reached the job execution time in the processing of S101 also in a node 2 to which CXL memory RM is not allocated (S104).

[0113] That is, in the memory determination process described with reference to Fig. 10 etc., the execution node 2 of the execution target job in which it is determined that at least one of the first execution time and the second execution time is not stored is determined to be the node 2 to which the CXL memory RM may be allocated (S13). Therefore, the job execution unit 115 executes the first execution of each execution target job in the node 2 to which the CXL memory RM may be allocated, for example, before the first execution of each execution target job in the node 2 to which the CXL memory RM is not allocated. Therefore, it can be determined that a case in which at least one of the first execution time and the second execution time is not stored as part of the execution time information 132 means that at least the first execution time is not stored.

[0114] Therefore, for example, when the job execution unit 115 determines that at least one of the first execution time and the second execution time is not stored, the job execution unit 115 determines whether each node 2 to which the CXL memory RM may be allocated is being used for other purposes. As a result, when it is determined that a node 2 that is not being used for other purposes exists, the job execution unit 115 executes the job to be executed also in the node 2 that is determined to exist.

[0115] This enables the information management unit 116 to obtain both the first execution time and the second execution time in response to the completion of execution of the job to be executed in each node 2, for example.

[0116] If it is determined that there is no node 2 that is not being used for other purposes, the job execution unit 115 may not perform the processes from S104 to S106, for example.

[0117] Returning to FIG. 13, the information management unit 116, for example, acquires the execution time (first execution time) of the job to be executed in the node 2 to which the CXL memory RM is not allocated, upon completion of the execution of the job to be executed in the node 2 to which the CXL memory RM is not allocated (S105).

[0118] That is, for example, the information management unit 116 acquires the execution time of the execution target job executed in the process of S104 in response to the completion of the execution of the execution target job executed in the process of S104.

[0119] Then, the information management unit 116 stores, for example, the execution time acquired in the process of S105 in the information storage area 130 as a part of the execution time information 132 (S106).

[0120] Thereafter, as shown in FIG. 14, for example, the job execution unit 115 refers to the execution time information 132 stored in the information storage area 130 and determines whether or not the execution time (second execution time) of the job to be executed in the processing of S102 when it is executed in node 2 to which the CXL memory RM may be allocated is stored (S111).

[0121] As a result, if it is determined that the execution time of the job to be executed in the processing of S102 when executed in node 2 to which CXL memory RM may be allocated is not stored (NO in S111), the information management unit 116 acquires, for example, the execution time (second execution time) of the job to be executed in node 2 to which CXL memory RM may be allocated upon completion of the execution of the job to be executed in node 2 to which CXL memory RM may be allocated (S112).

[0122] That is, for example, the information management unit 116 acquires the execution time of the execution target job executed in the process of S102 in response to the completion of the execution of the execution target job executed in the process of S102.

[0123] Then, the information management unit 116 stores, for example, the execution time acquired in the process of S112 in the information storage area 130 as a part of the execution time information 132 (S113).

[0124] On the other hand, if it is determined that the execution time when the job to be executed in the processing of S102 is executed in node 2 to which the CXL memory RM may be allocated is stored (YES in S111), the information management unit 116 does not perform the processing of S112 and S113, for example.

[0125] Then, the job executing unit 115 ends the job execution process, for example. The job executing unit 115 also ends the job execution process when it determines that both the execution time of the job executed in the process of S102 when it is executed in a node 2 to which the CXL memory RM is not allocated and the execution time of the job executed in the process of S102 when it is executed in a node 2 to which the CXL memory RM may be allocated are stored (YES in S103).

[0126] This enables the job scheduler 1a to accumulate, for example, the execution time corresponding to each job to be executed in the information storage area 130 as execution time information 132.

[0127] [Specific examples of the process of S106 and S113] Next, specific examples of the process of S106 and the process of S113 will be described.

[0128] For example, in the execution time information 132 shown in FIG. 18, in which "PG1" and "PG2" are set as the "Program ID" (information in the first line), no information is set for each of the "first execution time," "second execution time," and "access frequency."

[0129] Therefore, when the program IDs of the programs included in the job to be executed that has reached its job execution timing in the processing of S101 are PG1 and PG2, the information management unit 116 sets the first execution time acquired in the processing of S105 to the "first execution time" and sets the second execution time acquired in the processing of S112 to the "second execution time", as shown in the underlined portion of Fig. 20. In addition, the information management unit 116 sets the memory access frequency measured during the execution of the job to be executed that has reached its job execution timing in the processing of S101 to the "access frequency", as shown in the underlined portion of Fig. 20.

[0130] Also, for example, in the execution time information 132 shown in FIG. 18, in which "PG4" is set as the "program ID" (information on the fourth line), no information is set as the "first execution time".

[0131] Therefore, for example, when the program ID of a program included in a job to be executed when the job execution timing has arrived in the processing of S101 is PG4, the information management unit 116 sets the first execution time obtained in the processing of S105 to the "first execution time", for example, as shown in the underlined portion of FIG. 21.

[0132] [Memory redetermining process] Next, a specific example of a process of redetermining a memory group (node ​​2) for executing each job to be executed (hereinafter, also referred to as a memory redetermining process) among the job execution control processes will be described. Fig. 14 and Fig. 15 are sequence charts for explaining the memory redetermining process.

[0133] 14, the memory group redetermining unit 117 waits until the redetermining timing comes (NO in S31), for example. The redetermining timing may be every few hours, for example.

[0134] Then, when the time for redetermining arrives (YES in S31), the memory group redetermining unit 117 determines, for example, whether there is a waiting job determined to be in a node 2 (node ​​2 having the first memory group) to which the CXL memory RM is not assigned (S32).

[0135] As a result, when it is determined that there is a waiting job determined to be in a node 2 to which the CXL memory RM is not allocated (YES in S32), the memory group redetermining unit 117 identifies, for example, one of the waiting jobs (S33).

[0136] Next, the memory group redetermining unit 117 determines, for example, whether the waiting job identified in the processing of S33 is a job to be executed across multiple nodes 2 (hereinafter also referred to as a job to be executed that performs parallel processing) (S34).

[0137] Specifically, for example, if the amount of memory required to execute the waiting job identified in the processing of S33 exceeds the amount of memory installed in a node 2 (one node 2) to which CXL memory RM is not assigned, the memory group redetermining unit 117 determines that the waiting job identified in the processing of S33 is a job to be executed across multiple nodes 2.

[0138] As a result, if it is determined that the waiting job identified in the processing of S33 is a job to be executed across multiple nodes 2 (YES in S34), the memory group redetermination unit 117 determines, for example, the execution nodes 2 for the waiting job identified in the processing of S33 to be multiple nodes 2 to which CXL memory RM is not assigned and which have the same memory group configuration, or multiple nodes 2 to which CXL memory RM may be assigned and which have the same memory group configuration (S35).

[0139] Specifically, when the memory group re-determination unit 117 determines that it is possible to execute the execution node 2 of the waiting job identified in the processing of S33 on multiple nodes 2 (for example, nodes 2c and 2d described in Figure 1, etc.) to which CXL memory RM is not assigned and which have the same number of local memories LM, it determines the execution node 2 of the waiting job identified in the processing of S33 to be the multiple nodes 2 to which CXL memory RM is not assigned and which have the same number of local memories LM.

[0140] On the other hand, when the memory group re-determination unit 117 determines that the execution node 2 of the waiting job identified in the processing of S33 cannot be executed in a plurality of nodes 2 to which the CXL memory RM is not allocated and which have the same number of installed local memories LM, it determines the execution node 2 of the waiting job identified in the processing of S33 to be a node 2 to which the CXL memory RM may be allocated and which have the same number of installed local memories LM (for example, the nodes 2a and 2b described in FIG. 1, etc.). Furthermore, in this case, the job scheduler 1a allocates the same number of CXL memories RM to each of the execution nodes 2 of the waiting job identified in the processing of S33, for example.

[0141] That is, when a waiting job is executed in a plurality of nodes 2, for example, the processing results of each processing unit are aggregated in each node 2 for each predetermined processing unit. Therefore, if the configuration of the memory group in each node 2 is different, for example, a node 2 with a fast execution speed will have to wait for the completion of processing in a node 2 with a slow execution speed, and the execution performance of the job to be executed in the node 2 with the fast execution speed will decrease.

[0142] Therefore, for example, when the waiting job identified in the processing of S33 is a job that is executed across multiple nodes 2, the memory group redetermination unit 117 determines that the multiple execution nodes 2 that will execute the waiting job identified in the processing of S33 are multiple nodes 2 that have the same memory group configuration.

[0143] This allows the job scheduler 1a to, for example, suppress the occurrence of differences in processing speed among a plurality of execution nodes 2 that execute waiting jobs. Therefore, the job scheduler 1a can, for example, suppress the degradation of the execution performance of the job to be executed in each node 2.

[0144] Returning to FIG. 14, if it is determined that the waiting job identified in the processing of S33 is not a job to be executed across multiple nodes 2 (NO in S34), the memory group redetermining unit 117, as shown in FIG. 15, for example, refers to the execution time information 132 stored in the information storage area 130 and identifies the memory access frequency of the waiting job identified in the processing of S33 (S41).

[0145] Then, if it is determined that the memory access frequency identified in the processing of S41 is equal to or greater than a threshold value (hereinafter also referred to as the second threshold value) (YES in S42), the memory group redetermining unit 117, for example, determines (changes) the execution node 2 of the waiting job identified in the processing of S33 to a node 2 to which the CXL memory RM may be allocated (S43).

[0146] On the other hand, when it is determined that the memory access frequency identified in the process of S41 is less than the second threshold (NO in S42), the memory group redetermining unit 117 does not perform the process of S43, for example.

[0147] That is, when a waiting job with a high memory access frequency is executed in, for example, a node 2 to which a CXL memory RM is not allocated, it can be determined that, compared to an execution target job with a low memory access frequency, the waiting job with a high memory access frequency is likely to degrade the performance of other execution target jobs executed in the same node 2. Therefore, the memory group redetermining unit 117 suppresses the degradation in performance of each execution target job executed in a node 2 to which a CXL memory RM is not allocated, for example, by changing the execution node 2 of a waiting job with a memory access frequency equal to or higher than a threshold to a node 2 to which a CXL memory RM may be allocated.

[0148] Thereafter, the memory group redetermining unit 117 determines whether or not all waiting jobs have been identified in the process of S33 (S36). Note that the memory group redetermining unit 117 also performs the process of S36 after the process of S35, for example.

[0149] As a result, when it is determined that all the waiting jobs have not been identified in the process of S33 (NO in S36), the memory group redetermining unit 117 performs, for example, the processes from S33 onwards again.

[0150] On the other hand, if it is determined in the processing of S33 that all waiting jobs have been identified (YES in S36), or if it is determined that there are no waiting jobs (NO in S32), the memory group redetermining unit 117, for example, terminates the memory redetermining processing.

[0151] In addition, if there are no waiting jobs whose execution node 2 has been changed despite the memory redetermining process being performed, the memory group redetermining unit 117 may, for example, change the execution node 2 of at least some of the waiting jobs (for example, the waiting job that has the earliest execution order) to a node 2 to which the CXL memory RM may be assigned.

[0152] In this manner, the job scheduler 1a in this embodiment refers to the information storage area 130, which stores the first execution time of each job to be executed when the first memory group is used and the second execution time of each job to be executed when the second memory group is used, and identifies the first execution time and the second execution time corresponding to the job to be executed.

[0153] Then, the job scheduler 1a calculates, for example, a value indicating the degree of deviation between the identified first execution time and the second execution time. After that, the job scheduler 1a determines a specific memory group for executing the target job from the first memory group and the second memory group, for example, based on whether the calculated value satisfies a predetermined condition.

[0154] That is, the job scheduler 1a in this embodiment judges the magnitude of the deviation between the first execution time and the second execution time corresponding to each execution target job, for example, by referring to the first execution time and the second execution time stored in advance in the storage unit. Then, the job scheduler 1a estimates, for example, that an execution target job for which it has been determined that the deviation is large is an execution target job whose execution performance will be significantly degraded if executed in the node 2 to which the CXL memory RM is assigned (in other words, an execution target job that is not suitable for execution in the node 2 to which the CXL memory RM is assigned). Furthermore, the job scheduler 1a estimates, for example, that an execution target job for which it has not been determined that the deviation is large is an execution target job whose execution performance will not be significantly degraded even if executed in the node 2 to which the CXL memory RM is assigned (in other words, an execution target job that is suitable for execution in the node 2 to which the CXL memory RM is assigned).

[0155] Then, the job scheduler 1a decides to execute a job to be executed that is determined to be unsuitable for execution in a node 2 to which the CXL memory RM is allocated in a node 2 to which the CXL memory RM is not allocated (for example, node 2c or node 2d in FIG. 4). Also, the job scheduler 1a decides to execute a job to be executed that is determined to be suitable for execution in a node 2 to which the CXL memory RM is allocated in a node 2 to which the CXL memory RM may be allocated (for example, node 2a or node 2b in FIG. 4).

[0156] As a result, the job scheduler 1a in this embodiment can, for example, suppress the occurrence of waiting for execution of jobs to be executed in each node 2, and can also suppress a decrease in the execution performance of each job to be executed.

[0157] The above embodiment can be summarized as follows:

[0158] (Appendix 1) referring to a storage unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that requires longer access time than the first memory group is used, and identifying the first execution time and the second execution time corresponding to a specific job; calculating a value indicating a degree of deviation between the identified first execution time and the identified second execution time; determining a specific memory group for executing the specific job from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition; A job execution control program for causing a computer to execute a process.

[0159] (Appendix 2) In Appendix 1: the first memory group is a memory group consisting of one or more local memories; The second memory group is a memory group consisting of one or more local memories, or a memory group consisting of one or more local memories and one or more remote memories. 2. A job execution control program comprising:

[0160] (Appendix 3) In Appendix 2: Each of the one or more remote memories is a CXL (Compute Express Link) memory. 2. A job execution control program comprising:

[0161] (Appendix 4) In Appendix 1: the calculating step calculates a ratio of the first execution time to the second execution time as the value; In the process of determining, if the value is less than a first threshold, it is determined that the specific job is to be executed by using the first memory group, and if the value is equal to or greater than the first threshold, it is determined that the specific job is to be executed by using the second memory group. 2. A job execution control program comprising:

[0162] (Appendix 5) Appendix 1 further states: When the first execution time corresponding to the specific job is not stored in the storage unit, the specific job is executed using the first memory group to obtain the first execution time corresponding to the specific job; storing the acquired first execution time in the storage unit; A job execution control program for causing a computer to execute a process.

[0163] (Appendix 6) Appendix 1 further states: When the second execution time corresponding to the specific job is not stored in the storage unit, the specific job is executed using the second memory group to obtain the second execution time corresponding to the specific job; storing the acquired second execution time in the storage unit; A job execution control program for causing a computer to execute a process.

[0164] (Appendix 7) In Appendix 1: In the process of determining, when a plurality of memory groups are used when executing the specific job and the value is equal to or greater than a first threshold, determining to execute the specific job by using a plurality of the first memory groups, and when a plurality of memory groups are used when executing the specific job and the value is less than the first threshold, determining to execute the specific job by using a plurality of the second memory groups. 2. A job execution control program comprising:

[0165] (Appendix 8) In Appendix 1: The storage unit stores a frequency of access to the memory during execution of each job, and further When the first memory group is determined as the specific memory group in the determination process, the storage unit is referenced to identify the access frequency corresponding to the specific job; when the identified access frequency is equal to or greater than a second threshold, changing the identified memory group from the first memory group to the second memory group; A job execution control program for causing a computer to execute a process.

[0166] (Appendix 9) a time specifying unit that refers to a storage unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that requires longer access time than the first memory group is used, and specifies the first execution time and the second execution time corresponding to a specific job; a deviation calculation unit that calculates a value indicating a degree of deviation between the identified first execution time and the identified second execution time; a memory group determination unit that determines a specific memory group for executing the specific job from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition. 23. An information processing apparatus comprising:

[0167] (Appendix 10) referring to a storage unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that requires longer access time than the first memory group is used, and identifying the first execution time and the second execution time corresponding to a specific job; calculating a value indicating a degree of deviation between the identified first execution time and the identified second execution time; determining a specific memory group for executing the specific job from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition; A job execution control method, comprising the steps of: executing a process by a computer; [Explanation of symbols]

[0168] 1: Information processing device 1a: Job scheduler 2: Node 3: Switch 4: Memory case 10: Information processing system 101: CPU 102: Memory 103: I / O interface 104: Storage device 105: Bus 130: Information storage area

Claims

1. referring to a storage unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that requires longer access time than the first memory group is used, the first execution time and the second execution time corresponding to a specific job are identified; calculating a value indicating a degree of deviation between the identified first execution time and the identified second execution time; determining a specific memory group for executing the specific job from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition; A job execution control program for causing a computer to execute a process.

2. In claim 1, the first memory group is a memory group consisting of one or more local memories; The second memory group is a memory group consisting of one or more local memories, or a memory group consisting of one or more local memories and one or more remote memories.

2. A job execution control program comprising:

3. In claim 1, the calculating step includes calculating a ratio of the first execution time to the second execution time as the value; In the determination process, if the value is less than a first threshold, it is determined that the specific job is to be executed by using the first memory group, and if the value is equal to or greater than the first threshold, it is determined that the specific job is to be executed by using the second memory group.

2. A job execution control program comprising:

4. In claim 1, In the process of determining, when a plurality of memory groups are used when executing the specific job and the value is equal to or greater than a first threshold value, it is determined that the specific job is executed by using a plurality of the first memory groups, and when a plurality of memory groups are used when executing the specific job and the value is less than the first threshold value, it is determined that the specific job is executed by using a plurality of the second memory groups.

2. A job execution control program comprising:

5. In claim 1, The storage unit stores a frequency of access to the memory during execution of each job, and further When the first memory group is determined as the specific memory group in the determination process, the storage unit is referenced to identify the access frequency corresponding to the specific job; when the identified access frequency is equal to or greater than a second threshold, changing the identified memory group from the first memory group to the second memory group; A job execution control program for causing a computer to execute a process.

6. a time specifying unit that refers to a storage unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that requires a longer access time than the first memory group is used, and specifies the first execution time and the second execution time corresponding to a specific job; a deviation calculation unit that calculates a value indicating a degree of deviation between the identified first execution time and the identified second execution time; a memory group determination unit that determines a specific memory group for executing the specific job from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition.

23. An information processing apparatus comprising:

7. referring to a storage unit that stores a first execution time of each job when a first memory group is used and a second execution time of each job when a second memory group that requires longer access time than the first memory group is used, the first execution time and the second execution time corresponding to a specific job are identified; calculating a value indicating a degree of deviation between the identified first execution time and the identified second execution time; determining a specific memory group for executing the specific job from the first memory group and the second memory group based on whether the calculated value satisfies a predetermined condition; A job execution control method, comprising the steps of: executing a process by a computer;

Citation Information

Patent Citations

  • Microprocessor device provided with shared memory

    JP1996161225A