Job management program, job management method, and job management device
The job management program optimizes resource allocation in hybrid HPC systems by prioritizing job execution in on-premises or cloud environments based on expected completion times and costs, addressing the challenge of meeting deadlines and reducing costs in hybrid HPC environments.
Patent Information
- Application Number
- JP2023081840
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-17
- Publication Date
- 2024-11-28
AI Technical Summary
Existing job management systems in hybrid HPC environments struggle to meet job completion deadlines while optimizing resource allocation across on-premises and cloud environments, leading to increased costs and potential deadline misses.
A job management program that utilizes a hybrid system comprising on-premises and cloud computing devices, prioritizing job execution based on expected completion times and costs, reallocating jobs to the most cost-effective environment to meet deadlines.
The solution effectively reduces billing costs while ensuring job completion within specified deadlines by strategically utilizing on-premises and cloud resources, optimizing resource allocation and execution times.
Smart Images

Figure 2024165558000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a job management program, a job management method, and a job management apparatus. [Background technology]
[0002] In recent years, the size and amount of data handled has increased dramatically with the advancement of technology. Accordingly, attention has been focused on HPC (High Performance Computing) systems, which have high processing power for quickly processing data.
[0003] Because HPC systems are expensive, companies do not own many of them, and because they are not available for each user to use freely, each user often uses one HPC system individually. When an HPC system is used by multiple users individually, the system executes each submitted job sequentially.
[0004] In addition, there are hybrid HPC systems that process jobs using both an on-premise environment using a company's own servers and a cloud environment using servers located in an external paid cloud.
[0005] When a deadline exists for a business operation that uses the results of a job, the job is required to meet the job's end deadline, which is the execution completion deadline calculated backwards from the deadline. Conventionally, batch processing job schedulers are based on the idea of first-come, first-served. Therefore, in order to shorten delivery times on fixed computer equipment, priority was given to jobs that should be processed as quickly as possible, such as those with an approaching deadline, and jobs would overtake or be stopped.
[0006] In response to this, cloud services have become widespread in recent years, and as long as the usage fee is paid, it is now possible to execute other jobs promptly without stopping the running job. In a hybrid HPC system that uses both cloud services and on-premise computing facilities, it is desirable to achieve a good balance between the delivery date and the cloud service usage fee.
[0007] A management server technology has been proposed that optimizes the placement of resources for applications using the reducible cost of placement for each resource, including on-premise environments, private cloud environments, and public cloud environments. Also, a container execution management system has been proposed that manages tasks while referring to the operating status of both public clouds and on-premise data centers. [Prior art documents] [Patent documents]
[0008] [Patent Document 1] Patent Publication No. 2021-60745 [Patent Document 2] JP 2022-144518 A Summary of the Invention [Problem to be solved by the invention]
[0009] However, when deciding where to place resources, simply taking into account the possible cost savings, it may be difficult to meet work deadlines. Also, when managing tasks by referring to the work status in the cloud environment and on-premise environment, costs are not taken into account, making it difficult to properly place resources in actual operations.
[0010] The disclosed technique has been made in view of the above, and has an object to provide a job management program, a job management method, and a job management device that can reduce charges while meeting a job end deadline. [Means for solving the problem]
[0011] In one aspect of the job management program, the job management method, and the job management device disclosed in the present application, the job management program controls a hybrid system including an on-premise computing device operating in an on-premise environment and a cloud computing device operating in a billing cloud environment. The job management program causes the on-premise computing device to start execution of the first job when a first job, a second job, and a third job are executed in sequence, and when neither the second job nor the third job is completed by the respective end deadline times that are predetermined to be executed after the other job is executed after the first job is completed, if the expected execution time of the second job is shorter than the expected execution time of the third job, the job management program causes a computer to execute a process of allocating the second job to the cloud computing device and the third job to the on-premise computing device. Effect of the Invention
[0012] In one aspect, the present invention can reduce charges while meeting the deadline for completing a job. [Brief description of the drawings]
[0013] [Figure 1] FIG. 1 is a block diagram of a hybrid HPC system according to an embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a job management table. [Diagram 3] FIG. 3 is a diagram illustrating an example of a job execution history table. [Figure 4] FIG. 4 is a flowchart showing the entire job execution process. [Diagram 5] FIG. 5 is a flowchart of a resource selection and execution process performed by the job management device according to the first embodiment. [Figure 6] FIG. 6 is a flowchart of a billing amount calculation process performed by the job management device according to the first embodiment. [Figure 7]FIG. 7 illustrates an execution state in a case where jobs are executed sequentially in the order in which they were submitted and in a case where the job management device according to the first embodiment executes jobs. [Figure 8] FIG. 8 is a diagram illustrating an example of a job execution status by the job management device according to the second embodiment. [Figure 9A] FIG. 9A is a first flowchart of a resource selection and execution process by the job management device according to the third embodiment. [Figure 9B] FIG. 9B is a second flowchart of the resource selection and execution process by the job management device according to the third embodiment. [Figure 10] FIG. 10 is a diagram illustrating a comparison of execution results between the job management device according to the second embodiment and the job management device according to the third embodiment. [Figure 11] FIG. 11 is a diagram illustrating a comparison of execution status between the job management device according to the second embodiment and the job management device according to the third embodiment. [Figure 12] FIG. 12 is a diagram showing the hardware configuration of the job management device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0014] Hereinafter, embodiments of the job management program, the job management method, and the job management device disclosed in the present application will be described in detail with reference to the drawings. Note that the job management program, the job management method, and the job management device disclosed in the present application are not limited to the following embodiments. EXAMPLES
[0015] 1 is a block diagram of a hybrid HPC system according to an embodiment. The hybrid HPC system 1 includes a job management device 10, a user terminal device 20, an on-premise computing device 30, and a cloud computing device 40. The job management device 10 is connected to a plurality of user terminal devices 20, the on-premise computing devices 30, and the cloud computing devices 40 via a network.
[0016] There are a plurality of user terminal devices 20. For example, if the hybrid HPC system 1 is a system owned by a company, the user terminal devices 20 are terminal devices owned by each department and used by employees of each department. The user terminal devices 20 transmit jobs to be executed to the job management device 10. Although jobs can be executed by either the on-premise computing device 30 or the cloud computing device 40, it is preferable for a user requesting execution of a job to execute the job on the on-premise computing device 30, since the usage fee is low.
[0017] The on-premise computing device 30 is a high-performance computer that operates in an on-premise environment owned by an organization that owns the hybrid HPC system 1. The on-premise computing device 30 is, for example, a supercomputer. The on-premise computing device 30 may be one or more.
[0018] The on-premise computing device 30 executes the input job upon receiving a job execution command from the job management device 10. After the job execution, the on-premise computing device 30 transmits a notification of the completion of the execution of the job to the job management device 10 together with the execution result of the job.
[0019] The cloud computing device 40 is a high-performance computer that operates in a cloud environment outside the organization that owns the hybrid HPC system 1. The cloud computing device 40 is, for example, a supercomputer. The cloud computing device 40 may be one or more.
[0020] The cloud computing device 40 executes the input job upon receiving a job execution command from the job management device 10. After the job execution, the cloud computing device 40 transmits a notification of the completion of the job execution together with the execution result of the job to the job management device 10.
[0021] The job management device 10 is, for example, a server device. The job management device 10 receives jobs submitted from each user terminal device 20. Then, the job management device 10 schedules the submitted jobs to be executed one by one. At this time, the job management device 10 observes the maximum execution time and the end deadline time specified by the user and takes into consideration the usage fees of the on-premise computing device 30 and the cloud computing device 40. Then, the job management device 10 selects either the on-premise computing device 30 or the cloud computing device 40 as the device to execute each job. Then, the job management device 10 causes each job to be executed by the on-premise computing device 30 or the cloud computing device 40 according to the scheduling. Then, the job management device 10 transmits the execution results of the jobs to each user terminal device 20. Furthermore, the job management device 10 notifies each user terminal device 20 of the usage fee and charges the usage fee.
[0022] The following describes in detail the job management device 10. The job management device 10 includes a reception unit 11, a job management table 12, a job execution management unit 13, a job execution history table 14, and a billing amount calculation unit 15, as shown in FIG.
[0023] The reception unit 11 receives a job submitted by a user from the user terminal device 20. Here, the submitted job is accompanied by information on the maximum execution time and the end deadline time. The maximum execution time is the upper limit of the time to execute the submitted job, and when the maximum execution time has elapsed since the start of execution of the submitted job, the job execution ends, and the results obtained up to that point are regarded as the execution result. The end deadline time is the deadline for completing job execution required by the user, and is calculated backwards from the work delivery date, for example. The job management device 10 is required to complete the execution of the job by the end deadline time. In other words, the time obtained by subtracting the maximum execution time from the end deadline time is the execution start deadline time, which is the final deadline for starting execution of the job.
[0024] Next, the reception unit 11 registers the received job in the job management table 12. For example, the reception unit 11 adds a new job management record to the job management table 12, and registers the job submission time, the maximum execution time specified by the job submitter, and the end deadline time in the added job management record. Furthermore, the reception unit 11 calculates the execution start deadline time by subtracting the maximum execution time from the end deadline time. Next, the reception unit 11 registers the calculated execution start deadline time in the added job management record. The reception unit 11 also registers information of "waiting to start execution" in the execution status column of the added job management record, indicating that the job is not yet executed.
[0025] FIG. 2 is a diagram showing an example of a job management table. As shown in FIG. 2, the job management table 12 has a job management record represented by each row for each submitted job. The job submission time, maximum execution time, end deadline time, execution start deadline time, and execution status are registered in the job management record. The job management record is generated by the reception unit 11 when a job is submitted, and a job ID (Identifier) is assigned by the reception unit 11. Furthermore, the job submission time, maximum execution time, end deadline time, and execution start deadline time are registered in the job management record. The execution status in the job management table 12 is information indicating whether the job to which the job ID registered in the job management record is assigned is before execution, during execution, or after execution. When the job management record is first created, the execution status is registered as waiting to start by the reception unit 11, and the information is updated by the job execution management unit 13 according to the execution status of the job. In the following, a job whose execution state is waiting to start will be referred to as a "job waiting to start execution", a job whose execution state is currently being executed will be referred to as a "job currently being executed", and a job whose execution state is currently finished will be referred to as a "job that has finished being executed".
[0026] When the on-premise computing device 30 finishes executing a job and an available resource is generated, the job execution management unit 13 executes the following resource selection and execution process. If the on-premise computing device 30 is available with a job to be executed, any job is submitted to it, so if a job exists, the on-premise computing device 30 will execute the job continuously. The job execution management unit 13 checks the job management table 12, and obtains and lists information on jobs waiting to start execution registered in the job management table 12. If no jobs waiting to start execution are registered in the job management table 12, the job execution management unit 13 ends the job scheduling process.
[0027] When there is one job waiting to start execution, the job execution management unit 13 selects the on-premise computing device 30 as the device to execute the job. Then, the job execution management unit 13 submits the job to the on-premise computing device 30 and instructs the on-premise computing device 30 to execute the job.
[0028] Furthermore, when there are multiple jobs waiting to start execution, the job execution management unit 13 sorts the jobs waiting to start execution in the order in which they were submitted. Here, the first job among the sorted jobs waiting to start execution is called the "first execution candidate job," and the second job from the top is called the "second execution candidate job."
[0029] The job execution management unit 13 adds the maximum execution time of the first execution candidate job to the current time, and sets the result as the expected execution end time of the first execution candidate job. Next, the job execution management unit 13 compares the expected execution end time of the first execution candidate job with the execution start deadline time of the second execution candidate job. If the expected execution end time of the first execution candidate job is before the execution start deadline time of the second execution candidate job, the job execution management unit 13 selects the on-premise computing device 30 as a device that will execute the first execution candidate job. After that, the job execution management unit 13 submits the first execution candidate job to the on-premise computing device 30, and instructs the on-premise computing device 30 to execute the first execution candidate job.
[0030] If the expected execution end time of the first execution candidate job is later than the execution start deadline of the second execution candidate job, the job execution management unit 13 compares the maximum execution time of the first execution candidate job with the maximum execution time of the second execution candidate job.
[0031] If the maximum execution time of the first execution candidate job is equal to or longer than the maximum execution time of the second execution candidate job, the job execution management unit 13 selects the on-premise computing device 30 as a device for executing the first execution candidate job. Then, the job execution management unit 13 submits the first execution candidate job to the on-premise computing device 30 and instructs the on-premise computing device 30 to execute the first execution candidate job.
[0032] Conversely, if the maximum execution time of the first execution candidate job is shorter than the maximum execution time of the second execution candidate job, the job execution management unit 13 selects the cloud computing device 40 as a device for executing the first execution candidate job. Then, the job execution management unit 13 submits the first execution candidate job to the cloud computing device 40 and instructs the cloud computing device 40 to execute the first execution candidate job.
[0033] After that, the job execution management unit 13 waits until an available space is generated in the on-premise computing device 30. The job execution management unit 13 monitors whether or not the execution start deadline of the job in the execution start waiting state is reached during the waiting state. If the execution start deadline of the job in the execution start waiting state is not reached during the waiting state, the job execution management unit 13 executes the resource selection execution process described above. On the other hand, if the execution start deadline of the job in the execution start waiting state arrives during the waiting state, the job execution management unit 13 executes the following cloud usage execution process.
[0034] The job execution management unit 13 selects a cloud computing device 40 to execute the job waiting to start execution whose execution start deadline has arrived. Then, the job execution management unit 13 submits the job waiting to start execution whose execution start deadline has arrived to the cloud computing device 40, and instructs the cloud computing device 40 to execute the job.
[0035] The job execution management unit 13 executes the following process regardless of whether the job is executed by the on-premise computing device 30 or the cloud computing device 40. When the job execution starts, the job execution management unit 13 changes the execution state of the job management record corresponding to the job that has started to be executed in the job management table 12 to "in execution." Furthermore, the job execution management unit 13 adds a job execution history record to the job execution history table 14. Then, the job execution management unit 13 registers the job ID of the job that has started to be executed in the added job execution history record, and also registers the selected execution location, the execution ID assigned as the identification information of the job execution, and the execution start time.
[0036] Furthermore, when the job execution management unit 13 receives a notification of the end of job execution from the on-premise computing device 30 or the cloud computing device 40, it changes the execution state of the job management record corresponding to the job whose execution has been completed in the job management table 12 to "executed". Furthermore, the job execution management unit 13 registers the execution time in the job execution history record corresponding to the job whose execution has been completed in the job execution history table 14. Furthermore, when the job is executed by the cloud computing device 40, the job execution management unit 13 calculates the cloud service usage fee for the execution of the job based on a predetermined calculation method. Then, the job execution management unit 13 registers the cloud service usage fee in the job execution history record corresponding to the job whose execution has been completed.
[0037] 2 is a state 121 of the job management table 12 in which the job with job ID #1 is being executed, and the jobs with job IDs #2 and #3 are waiting to start execution. For example, hereinafter, the jobs with job IDs #1 to #3 are referred to as "jobs #1 to #3," respectively. When job #1 starts execution while jobs #2 and #3 are waiting to start execution, the job execution management unit 13 transitions the job management table 12 to state 121.
[0038] 2 is a state 122 in which job #1 has been completed and jobs #2 and #3 are being executed in the job management table 12. For example, when job #1 has been completed and jobs #2 to #3 are being executed, the job execution management unit 13 transitions the job management table 12 to the state 122.
[0039] 3 is a diagram showing an example of a job execution history table. When the on-premise computing device 30 starts execution of job #1 from a state where there is no job, the job execution management unit 13 adds a job execution history record of job #1 to the job execution history table 14. Furthermore, the job execution management unit 13 registers the on-premise computing device 30 as the execution location in the job execution history record of job #1, registers an execution ID and an execution start time, and transitions the job execution history table 14 to a state 141.
[0040] Here, the cloud service usage necessity flag in the job execution history table 14 is information indicating whether or not a cloud service could have been used to execute the job, and is registered by the billing amount calculation unit 15, which will be described later.
[0041] Furthermore, when job #2 is executed on the cloud computing device 40 and job #3 is executed on the on-premise computing device 30 from state 141, the job execution management unit 13 adds job execution history records for jobs #2 and #3. After that, when the execution of jobs #1 to #3 is completed, the job execution management unit 13 registers the respective execution times and cloud service fees, and transitions the job execution history table 14 to state 142.
[0042] The billing calculation unit 15 executes the following billing calculation process when the billing amount for the cloud service fee is calculated, for example, when all jobs have been executed and there are no jobs waiting to be executed. The billing calculation unit 15 calculates the total amount of the cloud service fee registered in the job execution history table 14.
[0043] Next, the billing calculation unit 15 sorts the jobs in the execution completed state registered in the job execution history table 14 in the order in which the jobs were submitted. Next, the billing calculation unit 15 obtains the job submission time of the top job in the sorted result from the job management table 12, and sets it as the initial value of the available time for the on-premise computing device 30. Next, the billing calculation unit 15 sets the initial value of the cloud service total usage time estimate to 0.
[0044] Next, the billing calculation unit 15 executes the following process for each job in the sorted jobs, starting from the beginning in sorted order. Here, the job to be processed is called the target job. The billing calculation unit 15 obtains the execution start deadline time of the target job from the job management table 12, and compares it with the available time of the on-premise computing device 30 at that time.
[0045] If the execution start deadline time of the target job is earlier than the available time of the on-premise computing device 30, the billing calculation unit 15 registers “Applicable” as the value of the cloud service usage requirement flag in the job execution history record of the target job in the job execution history table 14. Next, the billing calculation unit 15 adds the execution time of the target job to the estimated total cloud service usage time.
[0046] Conversely, if the execution start deadline time of the target job is after the available time of the on-premise computing device 30, "not applicable" is registered as the value of the cloud service usage requirement flag in the job execution history record of the target job in the job execution history table 14. Next, the billing amount calculation unit 15 adds the execution time of the target job to the available time of the on-premise computing device 30 to update the available time of the on-premise computing device 30.
[0047] For example, when the job execution history table 14 after the end of the execution of jobs #1 to #3 is in the state 142 of FIG. 3, the billing calculation unit 15 sets the job input time of job #1 as the initial value of the available time. Then, when the execution start deadline time of job #1 is later than the available time of the on-premise computing device 30, the billing calculation unit 15 registers "not applicable" as the value of the cloud service usage necessary flag of job #1 in the job execution history table 14. Furthermore, the billing calculation unit 15 adds the execution time of job #1 to 0, which is the initial value of the available time of the on-premise computing device 30, to update the available time of the on-premise computing device 30. Similarly, the billing calculation unit 15 registers the cloud service usage necessary flag and updates the available time of the on-premise computing device 30 for job #2. Next, the billing calculation unit 15 compares the execution start deadline time of job #3 with the updated available time of the on-premise computing device 30. If the execution start deadline time of job #3 is before the updated available time of the on-premise computing device 30, the billing calculation unit 15 registers "Applicable" as the value of the cloud service usage requirement flag of job #3 in the job execution history table 14. Furthermore, the billing calculation unit 15 adds the execution time of job #3 to the estimated total cloud service usage time. As a result, the billing calculation unit 15 transitions the job execution history table 14 to state 143 in FIG. 3.
[0048] After performing the above process for all sorted jobs, the billing calculation unit 15 divides the execution time by the estimated total cloud service usage time for each job whose cloud service usage requirement flag has the value "applicable." The billing calculation unit 15 then multiplies the total cloud service usage fee by the result of this division to calculate the billing amount for the cloud service usage fee for each job whose cloud service usage requirement flag has the value "applicable."
[0049] Thereafter, the billing calculation unit 15 notifies the calculated billing amount for cloud service usage to the user terminal device 2 that submitted the job for which the value of the cloud service usage necessity flag is set to "applicable." The billing calculation unit 15 also notifies the billing amount for usage of the on-premise computing device 30 to the user terminal device 2 that submitted the job for which the value of the cloud service usage necessity flag is set to "not applicable."
[0050] 4 is a flowchart showing the entire job execution process. Next, the overall flow of the job execution process in the hybrid HPC system 1 will be described with reference to FIG.
[0051] The job management device 10 determines whether or not an available space has been generated in the on-premise computing device 30 (step S1).
[0052] When a vacant space is generated in the on-premise computing device 30 (step S1: Yes), the job management device 10 executes a resource selection and execution process (step S2).
[0053] On the other hand, if there is no free space in the on-premise computing device 30 (step S1: No), the job management device 10 judges whether or not there is a job waiting to start execution whose execution start deadline has arrived (step S3). If there is no job waiting to start execution whose execution start deadline has arrived (step S3: No), the job execution process returns to step S1.
[0054] On the other hand, if there is a job waiting to start execution whose execution start deadline has arrived (step S3: Yes), the job management device 10 executes the cloud-utilizing execution process (step S4). Specifically, the job execution management unit 13 selects a cloud computing device 40 to execute the job waiting to start execution whose execution start deadline has arrived. Then, the job execution management unit 13 submits the job waiting to start execution whose execution start deadline has arrived to the cloud computing device 40, and instructs the cloud computing device 40 to execute the job.
[0055] Thereafter, the job management device 10 determines whether or not to end the operation based on the presence or absence of a shutdown instruction, etc. (Step S5) If the operation is to be continued (Step S5: No), the job execution process returns to Step S1.
[0056] On the other hand, if the operation is to be ended (step S5: Yes), the job management apparatus 10 ends the job execution process.
[0057] Fig. 5 is a flowchart of a resource selection execution process by the job management device according to the embodiment 1. The flow shown in Fig. 5 corresponds to an example of the process executed in step S2 in Fig. 4. Next, the flow of the resource selection execution process by the job management device 10 according to the embodiment will be described with reference to Fig. 5.
[0058] The job execution management unit 13 refers to the job management table 12 and lists jobs waiting to start execution (step S101).
[0059] Next, the job execution management unit 13 determines whether or not there is one or more jobs waiting to start execution (step S102). If there is no job waiting to start execution (step S102: No), the job execution management unit 13 ends the resource selection and execution process.
[0060] On the other hand, if there is one or more jobs waiting to start execution (step S102: Yes), the job execution management unit 13 judges whether there is one job waiting to start execution (step S103).
[0061] If there is one job waiting to start execution (step S103: Yes), the job execution management unit 13 selects the on-premise computing device 30 as a resource for executing the one job waiting to start execution (step S104).
[0062] Then, the job execution management unit 13 causes the on-premise computing device 30 to execute the job that is waiting to start execution (step S105). After that, the job execution management unit 13 ends the resource selection and execution process.
[0063] On the other hand, if there are a plurality of jobs waiting to start execution (step S103: No), the job execution management unit 13 sorts the jobs waiting to start execution in the order in which they were submitted (step S106).
[0064] Next, the job execution management unit 13 adds the maximum execution time of the first execution candidate job, which is the top job in the sorted result, to the current time and calculates the result as the expected execution end time of the first execution candidate job (step S107).
[0065] Next, the job execution management unit 13 determines whether the expected execution end time of the first execution candidate job is later than the execution start deadline time of the second execution candidate job (step S108).
[0066] If the expected end time of the first execution candidate job is later than the deadline for starting execution of the second execution candidate job (step S108: Yes), the job execution management unit 13 determines whether the maximum execution time of the first execution candidate job is longer than the maximum execution time of the second execution candidate job (step S109).
[0067] If the maximum execution time of the first execution candidate job is longer than the maximum execution time of the second execution candidate job (step S109: Yes), the job execution management unit 13 selects the cloud computing device 40 as a resource for executing the first execution candidate job (step S110).
[0068] Then, the job execution management unit 13 causes the first execution candidate job to be executed by the cloud computing device 40 (step S111). After that, the job execution management unit 13 ends the resource selection and execution process.
[0069] On the other hand, if the expected execution end time of the first execution candidate job is before the execution start deadline time of the second execution candidate job (step S108: No), the job execution management unit 13 executes the following process. Similarly, if the maximum execution time of the first execution candidate job is equal to or longer than the maximum execution time of the second execution candidate job (step S109: No), the job execution management unit 13 also executes the following process. The job execution management unit 13 selects the on-premise computing device 30 as a resource for executing the first execution candidate job (step S112).
[0070] Then, the job execution management unit 13 causes the first execution candidate job to be executed by the on-premise computing device 30 (step S113). After that, the job execution management unit 13 ends the resource selection and execution process.
[0071] 6 is a flowchart of the billing amount calculation process by the job management device 10 according to the embodiment 1. Next, the flow of the billing amount calculation process by the job management device 10 according to the embodiment will be described with reference to FIG.
[0072] The billing amount calculation unit 15 calculates the total amount of the cloud service usage fees registered in the job execution history table 14 (step S201).
[0073] Next, the billing amount calculation unit 15 sorts the jobs in the completed state registered in the job execution history table 14 in the order in which the jobs were submitted (step S202).
[0074] Next, the billing calculation unit 15 obtains the job input time of the top job in the sorted result from the job management table 12, and sets it as the initial value of the available start time of the on-premise computing device 30 (step S203).
[0075] Next, the billing amount calculation unit 15 sets the estimated total cloud service usage time to an initial value of 0 (step S204).
[0076] Next, the billing calculation unit 15 assigns consecutive numbers from 1 to the top job in the sorting result, in order, up to the last job. Here, it is assumed that the last job is assigned N. Then, the billing calculation unit 15 sets n to 1 (step S205).
[0077] Next, the billing amount calculation unit 15 judges whether n is greater than N (step S206).
[0078] If n is equal to or smaller than N (step S206: No), the billing amount calculation unit 15 selects the job to which the number n is assigned as the target job (step S207).
[0079] The execution start deadline of the target job is obtained from the job management table 12, and it is determined whether or not it is before the time when the on-premise computing device 30 can start being used (step S208).
[0080] If the execution start deadline of the target job is before the available time of the on-premise computing device 30 (step S208: Yes), the billing calculation unit 15 sets the value of the cloud service usage necessity flag of the target job to "yes" (step S209).
[0081] Next, the billing amount calculation unit 15 updates the estimated total cloud service usage time by adding the execution time of the job to which the number n is assigned (step S210).
[0082] On the other hand, if the execution start deadline time of the target job is later than the available time of the on-premise computing device 30 (step S208: No), the billing amount calculation unit 15 sets the value of the cloud service usage requirement flag of the target job to “not applicable” (step S211).
[0083] Next, the billing amount calculation unit 15 adds the execution time of the target job to the available time of the on-premise computing device 30 to update the available time of the on-premise computing device 30 (step S212).
[0084] After that, the billing amount calculation unit 15 increments n by 1 (step S213). Then, the billing amount calculation process returns to step S206. This allows the billing amount calculation unit 15 to set the value of the cloud service usage requirement flag for the sorted jobs from the first job to the last job.
[0085] On the other hand, if n is greater than N (step S206: Yes), the billing amount calculation unit 15 sets n to 1 (step S214).
[0086] Next, the billing amount calculation unit 15 selects the job to which the number n is assigned as the target job (step S215).
[0087] Next, the billing amount calculation unit 15 determines whether the value of the cloud service usage necessity flag of the target job is "applicable" (step S216). If the value of the cloud service usage necessity flag of the target job is "not applicable" (step S216: No), the billing amount calculation process proceeds to step S218.
[0088] On the other hand, if the value of the cloud service usage necessity flag of the target job is "corresponding" (step S216: Yes), the billing calculation unit 15 divides the execution time by the estimated total cloud service usage time. Then, the billing calculation unit 15 multiplies the total cloud service usage fee by the result of the division to calculate the billing amount for the cloud service usage fee of the target job (step S217).
[0089] Thereafter, the billing amount calculation unit 15 judges whether n has reached N (step S218). If n is less than N (step S218: No), the billing amount calculation unit 15 increments n by 1 (step S219). Then, the billing amount calculation process returns to step S215.
[0090] On the other hand, if n reaches N (step S218: Yes), the billing calculation unit 15 notifies the user terminal device 2 that is the sender of a job for which the cloud service usage necessary flag has a value of "applicable" of the cloud service usage fee billing amount for that job as the usage fee billing amount. Also, the billing calculation unit 15 notifies the user terminal device 2 that is the sender of a job for which the cloud service usage necessary flag has a value of "not applicable" of the usage fee billing amount for the on-premise computing device 30 (step S220).
[0091] 7 is a diagram showing the execution status when jobs are executed sequentially in the order of submission and when the job management device according to the first embodiment executes jobs. In graphs 201 and 202, the horizontal axis represents time and the vertical axis represents instances. End deadline times 211 to 213 are the end deadline times of jobs #1 to #3, respectively. Here, a case will be described in which one on-premise computing device 30 executes jobs one by one.
[0092] Graph 201 shows the execution results when jobs are executed sequentially in the order in which they were submitted. The execution of jobs with diagonal lines indicates job execution by the cloud computing device 40. In this case, the on-premise computing device 30 starts executing job #2 after finishing execution of job #1. Then, since the execution start deadline time for job #3 arrives while job #2 is being executed, job #3 is executed by the cloud computing device 40.
[0093] In contrast, graph 202 shows the results of job execution by the job management device 10 according to the first embodiment. In this case, after the execution of job #1 is completed, the execution time of job #3 is longer than the execution time of job #2, so job #2 is executed by the cloud computing device 40. Then, job #3 is executed by the on-premise computing device 30.
[0094] In this way, the job management device 10 according to the present embodiment executes jobs with short execution times on the cloud computing device 40, and executes jobs with long execution times on the on-premise computing device 30. This makes it possible to reduce overall charges in the hybrid HPC system 1.
[0095] As described above, the job management device according to the present embodiment executes a job in an on-premise environment if the expected execution end time of the job is earlier than the execution start deadline time of the next job. Also, if the maximum execution time of the job is equal to or longer than the maximum execution time of the next job, the job management device executes the job in an on-premise environment. Conversely, if the expected execution end time of the job is after the execution start deadline time of the next job and the maximum execution time is shorter than the maximum execution time of the next job, the job management device executes the job in a cloud environment. Furthermore, if the execution start deadline time calculated from the specified end deadline time of a specific job arrives when there is no free space in the on-premise environment, the job management device executes the job in a cloud environment.
[0096] This allows jobs with as short an execution time as possible to be executed in the cloud environment while keeping to the specified deadline for each job, thereby reducing charges while keeping to the deadline for each job.
[0097] (Modification) In the first embodiment, the job management device 10 uses the maximum execution time to determine whether to execute a job on the on-premise computing device 30 or the cloud computing device 40. However, the maximum execution time is a time that is overestimated in order to complete a job, and the execution usually ends in a shorter time than the maximum execution time. Therefore, the job management device 10 according to this modification predicts the execution time of each job from the execution results of past jobs and determines whether to execute the job on the on-premise computing device 30 or the cloud computing device 40. Details of the on-premise computing device 30 according to this modification will be described below.
[0098] The job execution management unit 13 according to this modification has in advance execution time prediction coefficients calculated from the results of many jobs executed in the past, according to the characteristics of the jobs, such as the type of calculation, the number of calculations, etc.
[0099] The job execution management unit 13 identifies an execution time prediction coefficient according to the characteristics of the first execution candidate job. Then, the job execution management unit 13 calculates an expected execution time of the first execution candidate job by multiplying the maximum execution time of the first execution candidate job by the execution time prediction coefficient of the first execution candidate job. Next, the job execution management unit 13 adds the expected execution time of the first execution candidate job to the current time, and sets the result as the expected execution end time of the first execution candidate job.
[0100] Next, the job execution management unit 13 compares the predicted execution end time of the first execution candidate job with the execution start deadline time of the second execution candidate job. If the predicted execution end time of the first execution candidate job is before the execution start deadline time of the second execution candidate job, the job execution management unit 13 selects the on-premise computing device 30 as a device that will execute the first execution candidate job. After that, the job execution management unit 13 submits the first execution candidate job to the on-premise computing device 30 and instructs the on-premise computing device 30 to execute the first execution candidate job.
[0101] Conversely, if the predicted execution end time of the first execution candidate job is later than the execution start deadline time of the second execution candidate job, the job execution management unit 13 identifies an execution time prediction coefficient according to the characteristics of the second execution candidate job.Then, the job execution management unit 13 multiplies the maximum execution time of the second execution candidate job by the execution time prediction coefficient of the second execution candidate job to calculate the predicted execution time of the second execution candidate job.Next, the job execution management unit 13 compares the predicted execution time of the first execution candidate job with the predicted execution time of the second execution candidate job.
[0102] If the expected execution time of the first execution candidate job is equal to or longer than the expected execution time of the second execution candidate job, the job execution management unit 13 selects the on-premise computing device 30 as a device that will execute the first execution candidate job.
[0103] Conversely, if the expected execution time of the first execution candidate job is shorter than the expected execution time of the second execution candidate job, the job execution management unit 13 selects the cloud computing device 40 as the device that will execute the first execution candidate job.
[0104] As described above, the job management device according to this modified example predicts the actual execution time of each job from the maximum execution time based on the execution results of the past jobs, and determines whether to execute the job in an on-premise environment or a cloud environment. The maximum execution time is estimated with a margin of error, and there are many cases where the maximum execution time differs greatly from the actual execution time. Therefore, the execution time predicted based on the execution results of the past jobs can be considered to be a more realistic value, and the execution environment of the job can be selected more appropriately. Therefore, it is possible to more appropriately reduce charges while meeting the job end deadline. EXAMPLES
[0105] Next, a second embodiment will be described. A hybrid HPC system 1 according to this embodiment is also shown in FIG. 1. A job management device 10 according to this embodiment causes an on-premise computing device 30 to execute a job whose execution start deadline is as close as possible and whose maximum execution time is as long as possible. Then, when the on-premise computing device 30 becomes capable of executing a job waiting to start execution by the execution start deadline, the job management device 10 causes the on-premise computing device 30 to execute the job. Details of the selection of a job execution environment by the job management device 10 according to this embodiment will be described below. In the following description, the functions of each unit similar to those of the first embodiment will be omitted.
[0106] When the on-premise computing device 30 finishes executing a job and an available resource is generated, the job execution management unit 13 executes the following resource selection and execution process. The job execution management unit 13 checks the job management table 12, and obtains and lists information on jobs waiting to start execution registered in the job management table 12. If no jobs waiting to start execution are registered in the job management table 12, the job execution management unit 13 ends the job scheduling process.
[0107] When there is one job waiting to start execution, the job execution management unit 13 selects the on-premise computing device 30 as the device to execute the job. Then, the job execution management unit 13 submits the job to the on-premise computing device 30 and instructs the on-premise computing device 30 to execute the job.
[0108] On the other hand, when there are multiple jobs waiting to start execution, the job execution management unit 13 sorts the acquired jobs waiting to start execution in ascending order of execution start deadline time and descending order of maximum execution time. Here, the first job among the sorted jobs waiting to start execution is called the "first execution candidate job," and the second job from the top is called the "second execution candidate job." After this, the job execution management unit 13 performs the same process as in the first embodiment.
[0109] That is, the job execution management unit 13 according to this embodiment differs from the job execution management unit 13 of the first embodiment in that it sorts the acquired waiting jobs in ascending order of execution start deadline time and descending order of maximum execution time. By sorting in this manner, the job execution management unit 13 executes the jobs with the earliest execution start deadline time on the on-premise computing device 30. Furthermore, the job execution management unit 13 executes the jobs with the latest execution start deadline time on the on-premise computing device 30 after the execution of the jobs with the earlier execution start deadline time is completed, or executes them on the cloud computing device 40 when the execution start deadline time arrives before that.
[0110] 8 is a diagram showing an example of the execution status of a job by a job management device according to the second embodiment. In graphs 203 and 204, the horizontal axis represents time and the vertical axis represents instances. End deadline times 211-213 are the end deadline times of jobs #1-#3, respectively. Here, a case will be described in which one on-premise computing device 30 executes one job at a time. The execution of a job marked with a diagonal line pattern represents the execution of a job by a cloud computing device 40.
[0111] Graph 203 shows the execution result when no free space is generated in the on-premise computing device 30 by the execution start deadline of the next job to be executed. After the execution of job #1 is completed, the job management device 10 causes the on-premise computing device 30 to execute job #3 because the execution start deadline of job #2 is later than the execution start deadline of job #3. Thereafter, when the execution start deadline of job #2 arrives while job #3 is being executed, the job management device 10 causes the cloud computing device 40 to execute job #2, as shown in graph 203.
[0112] In contrast, when the execution of job #3 ends before the execution start deadline of job #2, the job management device 10 performs resource selection and execution processing for job #2 at that time. In this case, since the on-premise computing device 30 is available and there is only job #2, the job management device 10 selects the on-premise computing device 30 as a resource for executing job #2. Then, the job management device 10 causes the on-premise computing device 30 to execute job #2, as shown in graph 204.
[0113] In the above description, the job execution management unit 13 selects the execution environment for each job using the maximum execution time. However, instead of the maximum execution time, the job execution management unit 13 may use the result of multiplying the maximum execution time by the execution time prediction coefficient of the target job.
[0114] As described above, the job management device according to this embodiment waits until the execution start deadline of each job, and if the job can be executed in the on-premise environment by that execution start deadline, executes the job in the on-premise environment, otherwise executes the job in the cloud environment. This allows more jobs to be executed in the on-premise environment, and makes it possible to appropriately reduce charges while keeping to the job end deadline. EXAMPLES
[0115] Next, a third embodiment will be described. A hybrid HPC system 1 according to this embodiment is also shown in FIG. 1. A job management device 10 according to this embodiment compares two or more jobs and determines a job to be executed by an on-premise computing device 30 so as to shorten the total usage time of the cloud computing device 40 as much as possible. The details of the selection of the execution environment of a job by the job management device 10 according to this embodiment will be described below. In the following explanation, the explanation of the functions of each unit similar to those of the first embodiment will be omitted. Here, a case will be explained in which there are multiple on-premise computing devices 30, each of which executes one job at a time sequentially.
[0116] When the on-premise computing device 30 finishes executing a job and an available resource is generated, the job execution management unit 13 executes the following resource selection and execution process. The job execution management unit 13 checks the job management table 12, and obtains and lists information on jobs waiting to start execution registered in the job management table 12. If no jobs waiting to start execution are registered in the job management table 12, the job execution management unit 13 ends the job scheduling process.
[0117] When there is one job waiting to start execution, the job execution management unit 13 selects the on-premise computing device 30 as the device to execute the job. Then, the job execution management unit 13 submits the job to the on-premise computing device 30 and instructs the on-premise computing device 30 to execute the job.
[0118] On the other hand, when there are a plurality of waiting jobs, the job execution management unit 13 sorts the acquired waiting jobs in ascending order of execution start deadline time and descending order of maximum execution time.
[0119] The job execution management unit 13 adds the maximum execution time of the top job in the sorted result to the current time, and sets the result as the expected execution end time of the top job. Also, for each job being executed in the on-premise computing device 30, the job execution management unit 13 adds the maximum execution time to the execution start time for each job being executed, and sets the sum as the expected execution end time of the job being executed. Then, the job execution management unit 13 sets the fastest expected execution end time among the expected execution end times of the top job and all jobs being executed in the on-premise computing device 30 as the expected time when the on-premise computing device 30 becomes available.
[0120] Next, the job execution management unit 13 extracts jobs whose execution start deadline time is earlier than the expected time when availability occurs in the on-premise computing device 30 from among the jobs in the sorted result. Then, the job execution management unit 13 sets the extracted jobs as the first to Nth on-premise execution candidate jobs in the order of sorting.
[0121] Next, the job execution management unit 13 selects a target on-premise execution candidate job from each of the first to Nth on-premise execution candidate jobs in order, and executes the following resource selection and execution process. The job execution management unit 13 adds the maximum execution time of the target on-premise execution candidate job to the current time, and sets the result as the initial value of the expected time when the on-premise computing device 30 can start using the job. The job execution management unit 13 also sets the initial value of the estimated cloud service total usage time to 0.
[0122] Next, the job execution management unit 13 selects jobs one by one in numerical order from among the first to Nth on-premise execution candidate jobs other than the target on-premise execution candidate job, and performs the following processing in order. Hereinafter, the job selected by the job execution management unit 13 is called a selected job. If the execution start deadline of the selected job is earlier than the expected time when the on-premise computing device 30 is available, the job execution management unit 13 adds the maximum execution time of the selected job to the estimated cloud service total usage time. On the other hand, if the execution start deadline of the selected job is after the expected time when the on-premise computing device 30 is available, the job execution management unit 13 adds the maximum execution time of the selected job to the expected time when the on-premise computing device 30 is available.
[0123] As a result, the job execution management unit 13 calculates the estimated cloud service total usage time value for the N target on-premise execution candidate jobs. Then, the job execution management unit 13 selects the on-premise execution candidate job for which the estimated cloud service total usage time value is the smallest. Then, the job execution management unit 13 selects the on-premise computing device 30 as a resource for executing the selected on-premise execution candidate job. After that, the job execution management unit 13 submits the selected on-premise execution candidate job to the on-premise computing device 30, and causes the on-premise computing device 30 to execute the selected on-premise execution candidate job.
[0124] Fig. 9A is a first flowchart of the resource selection execution process by the job management device according to the embodiment 3. Fig. 9B is a second flowchart of the resource selection execution process by the job management device according to the embodiment 3. Next, the flow of the resource selection execution process by the job management device 10 according to the embodiment will be described with reference to Figs. 9A and 9B.
[0125] The job execution management unit 13 refers to the job management table 12 and lists jobs waiting to start execution (step S301).
[0126] Next, the job execution management unit 13 judges whether or not there is one or more jobs waiting to start execution (step S302). If there is no job waiting to start execution (step S302: No), the job execution management unit 13 ends the resource selection and execution process.
[0127] On the other hand, if there is one or more jobs waiting to start execution (step S302: Yes), the job execution management unit 13 judges whether there is one job waiting to start execution (step S303).
[0128] If there is one job waiting to start execution (step S303: Yes), the job execution management unit 13 selects the on-premise computing device 30 as a resource for executing the one job waiting to start execution (step S304).
[0129] Then, the job execution management unit 13 causes the on-premise computing device 30 to execute the job that is waiting to start execution (step S305). After that, the job execution management unit 13 ends the resource selection and execution process.
[0130] On the other hand, if there are multiple jobs waiting to start execution (step S303: No), the job execution management unit 13 sorts the jobs waiting to start execution in ascending order of execution start deadline time and in descending order of maximum execution time (step S306).
[0131] Next, the job execution management unit 13 adds the maximum execution time of the top job in the sorted result to the current time to calculate the expected execution end time of the top job (step S307).
[0132] Furthermore, the job execution management unit 13 calculates the expected execution end time for each of the jobs being executed in the on-premise computing device 30 by adding the maximum execution time to the execution start time for each of the jobs being executed (step S308).
[0133] Next, the job execution management unit 13 determines the earliest expected execution end time among the expected execution end times of the leading job and all jobs being executed in the on-premise computing device 30 as the expected time when the on-premise computing device 30 will become available (step S309).
[0134] Next, the job execution management unit 13 extracts jobs whose execution start deadline time is earlier than the expected time when availability occurs in the on-premise computing device 30 from among the jobs in the sorted result. Then, the job execution management unit 13 determines the extracted jobs as the first to Nth on-premise execution candidate jobs in the order of sorting (step S310).
[0135] Next, the job execution management unit 13 sets n=1 (step S311).
[0136] Next, the job execution management unit 13 sets the n-th on-premise execution candidate job as a target on-premise execution candidate job (step S312).
[0137] Next, the job execution management unit 13 adds the maximum execution time of the target on-premise execution candidate job to the current time and sets the result as an initial value of the expected time when the on-premise computing device 30 can start using (step S313).
[0138] Furthermore, the job execution management unit 13 sets the initial value of the estimated total cloud service usage time to 0 (step S314).
[0139] Next, the job execution management unit 13 selects, as a selected job, one of the first to N-th on-premise execution candidate jobs that is an unselected job with the smallest number other than the n-th on-premise execution candidate job (step S315).
[0140] The job execution management unit 13 determines whether the execution start deadline time of the selected job is earlier than the expected time when the on-premise computing device 30 becomes available (step S316).
[0141] If the execution start deadline of the selected job is earlier than the expected time when the on-premise computing device 30 becomes available (step S316: Yes), the maximum execution time of the selected job is added to the estimated total cloud service usage time (step S317).
[0142] On the other hand, if the execution start deadline of the selected job is after the expected time when the on-premise computing device 30 becomes available (step S316: No), the job execution management unit 13 adds the maximum execution time of the selected job to the expected time when the on-premise computing device 30 becomes available (step S318).
[0143] After that, the job execution management unit 13 judges whether all the on-premise execution candidate jobs other than the n-th on-premise execution candidate job have been considered (step S319). If there are any on-premise execution candidate jobs that have not been considered (step S319: No), the resource selection and execution process returns to step S315.
[0144] On the other hand, if all on-premise execution candidate jobs other than the n-th on-premise execution candidate job have been considered (step S319: Yes), the job execution management unit 13 proceeds to the next process. The job execution management unit 13 determines whether n=N, that is, whether all on-premise execution candidate jobs have been considered as target on-premise execution candidate jobs (step S320).
[0145] If n=N is not satisfied (step S320: No), the job execution management unit 13 increments n by 1 (step S321). After that, the resource selection and execution process returns to step S312.
[0146] On the other hand, if n=N (step S320: Yes), the job execution management unit 13 An on-premise execution candidate job that has the smallest estimated total cloud service usage time is selected (step S322).
[0147] After that, the job execution management unit 13 causes the on-premise computing device 30 to execute the selected on-premise execution candidate job (step S323).
[0148] 10 is a diagram showing a comparison of execution results between the job management device according to the second embodiment and the job management device according to the third embodiment. Here, a case where jobs #11 to #13 registered in the job management table 12 shown in FIG. 10 are executed will be described.
[0149] The job execution history table 14 showing the execution result when a job is executed using the job management device 10 according to the second embodiment is as shown in state 301. The job execution history table 14 showing the execution result when a job is executed using the job management device 10 according to the third embodiment is as shown in state 302. In this case, the total amount of cloud service usage fees is reduced from 16.4 to 15.2 when the job is executed using the job management device 10 according to the third embodiment, compared to when the job is executed using the job management device 10 according to the second embodiment. In other words, charges can be more appropriately reduced.
[0150] 11 is a diagram showing a comparison of execution status between the job management device according to the second embodiment and the job management device according to the third embodiment. In the graphs 303 and 304, the horizontal axis represents time and the vertical axis represents instances. End deadline times 311 to 313 are the end deadline times of jobs #11 to #13, respectively. The execution of jobs marked with a diagonal line pattern represents job execution by the cloud computing device 40.
[0151] Graph 303 shows the execution result by the job management apparatus 10 according to the second embodiment. In this case, the job management apparatus 10 according to the second embodiment causes the on-premise computing device 30 to execute job #13, which has the earliest execution start deadline time and the longest maximum execution time. Then, since the execution start deadline times of jobs #11 and #12 arrive while job #13 is being executed, the job management apparatus 10 according to the second embodiment causes the cloud computing device 40 to execute jobs #11 and #12.
[0152] In contrast, the graph 304 shows the execution result by the job management apparatus 10 according to the third embodiment. In this case, the job management apparatus 10 according to the third embodiment selects job #12 as a job having an execution start deadline as early as possible and a maximum execution time as long as possible, and executes it on the on-premise computing device 30. Since the execution start deadline of job #13 arrives while job #12 is being executed, the job management apparatus 10 according to the third embodiment executes job #13 on the cloud computing device 40. Since the execution start deadline of job #11 does not arrive before the execution of job #12 ends, the job management apparatus 10 according to the third embodiment can execute job #11 on the on-premise computing device 30. The job management apparatus 10 according to the third embodiment can finish the execution of all of jobs #11 to #13 by the end deadline.
[0153] In the above description, the job execution management unit 13 selects the execution environment for each job using the maximum execution time. However, instead of the maximum execution time, the job execution management unit 13 may use the result of multiplying the maximum execution time by the execution time prediction coefficient of the target job.
[0154] As described above, the job management device according to the present embodiment considers two or more jobs together and executes in an on-premise environment a job that has the earliest possible execution start deadline and the longest possible maximum execution time, and that shortens the total cloud service usage time as much as possible. This allows more jobs to be executed in the on-premise environment, and makes it possible to more appropriately reduce charges while keeping to the job end deadlines.
[0155] (Hardware configuration) 12 is a diagram showing a hardware configuration of the job management device 10. Next, an example of a hardware configuration for realizing each function of the job management device 10 will be described with reference to FIG.
[0156] 12, the job management apparatus 10 includes, for example, a CPU (Central Processing Unit) 91, a memory 92, a hard disk 93, and a network interface 94. The CPU 91 is connected to the memory 92, the hard disk 93, and the network interface 94 via a bus.
[0157] The network interface 94 is an interface for communication between the job management device 10 and an external device. The network interface 94 relays communication between the CPU 91 and the user terminal device 20, the on-premise computing device 30, and the cloud computing device 40, for example.
[0158] The hard disk 93 is an auxiliary storage device. The hard disk 93 stores the job management table 12 and the job execution history table 14 shown in Fig. 1. The hard disk 93 also stores programs that realize the functions of the reception unit 11, the job execution management unit 13, and the billing amount calculation unit 15 shown in Fig. 1.
[0159] The memory 92 is a main storage device, and may be, for example, a dynamic random access memory (DRAM).
[0160] The CPU 91 reads out various programs from the hard disk 93, and loads and executes them in the memory 92. In this way, the CPU 91 realizes the functions of the reception unit 11, the job execution management unit 13, and the billing amount calculation unit 15 illustrated in FIG. [Explanation of symbols]
[0161] 1 Hybrid HPC System 10 Job management device 11 Reception 12 Job Management Table 13 Job Execution Management Unit 14 Job execution history table 15 Billing Amount Calculation Department 20 User terminal device 30 On-Premise Computing Devices 40 Cloud Computing Device
Claims
1. A job management program for controlling a hybrid system including an on-premise computing device operating in an on-premise environment and a cloud computing device operating in a billing cloud environment, comprising: When the first job, the second job, and the third job are executed in this order, Initiating execution of the first job on the on-premise computing device; When neither the second job nor the third job is completed by a respective end deadline time that is predetermined to be executed after the other job is executed after the first job is completed, if an expected execution time of the second job is shorter than an expected execution time of the third job, the second job is assigned to the cloud computing device and the third job is assigned to the on-premise computing device. A job management program that causes a computer to execute a process.
2. causing the on-premise computing device to execute the third job after completion of execution of the first job; When the execution of the third job is completed by an execution start deadline time for finishing the second job by an end deadline time of the second job, the on-premise computing device is caused to execute the second job after the execution of the third job is completed.
2. The job management program according to claim 1, further causing the computer to execute a process.
3. 2. The job management program according to claim 1, further comprising: acquiring a maximum execution time and a completion deadline time for each of the first job, the second job, and the third job; and causing the computer to execute a process for determining, based on the maximum execution time and the predicted execution time, whether or not the execution of the third job will be completed by a predetermined completion deadline time for the third job if the execution of the third job is started after the execution of the first job is completed.
4. 4. The job management program according to claim 3, further comprising: calculating a predicted execution time for each of the first job, the second job, and the third job by multiplying the maximum execution time by an execution time prediction coefficient representing a ratio between the maximum execution time based on past execution results of the same type of job and an actual execution result; and causing the computer to execute a process of determining, based on the calculated predicted execution time, whether or not execution of the third job will be completed by a predetermined deadline for ending the third job if execution of the third job is started after completion of execution of the first job.
5. obtaining a maximum execution time and a deadline for completion of each of the second job and the third job; calculating an execution start deadline time for each of the second job and the third job by subtracting the maximum execution time from the end deadline time; acquiring execution times, which are execution results, for the second job and the third job, Identifying jobs to be executed on the cloud computing device when the second job and the third job are executed in order based on the execution start deadline time, and calculating the usage fees of the cloud computing device for each of the second job and the third job by prorating the total amount of the usage fees of the cloud computing device based on the execution time of the identified jobs.
2. The job management program according to claim 1, further causing the computer to execute a process.
6. A job management method in a hybrid system including an on-premise computing device operating in an on-premise environment and a cloud computing device operating in a billing cloud environment, comprising: A job management device When the first job, the second job, and the third job are executed in this order, Initiating execution of the first job on the on-premise computing device; When neither the second job nor the third job is completed by a respective end deadline time that is predetermined to be executed after the other job is executed after the first job is completed, if the expected execution time of the second job is shorter than the expected execution time of the third job, the second job is assigned to the cloud computing device, and the third job is assigned to an on-premise computing device.
23. A job management method comprising:
7. A hybrid system including an on-premise computing device operating in an on-premise environment, a cloud computing device operating in a billing cloud environment, and a job management device, The job management device includes: a reception unit that sequentially receives the first job, the second job, and the third job; a job execution management unit that starts execution of the first job in the on-premise computing device when the first job, the second job, and the third job are sequentially executed in the order of reception by the reception unit, and when neither the second job nor the third job is completed by a respective end deadline time that is predetermined to be executed after the other job is executed after the first job is completed, assigns the second job to the cloud computing device if an expected execution time of the second job is shorter than an expected execution time of the third job, and assigns the third job to the on-premise computing device; A job management device comprising:
Citation Information
Patent Citations
Assisting device and method
JP2021060745A
Operation method, and operation instruction system
JP2022144518A
