JP2004227188A - Job swap method, job management device and job management program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To attain the reduction of overhead required for job swap and to effectively utilize memory resources. <P>SOLUTION: The swap priority of a partially swapped-out job is changed to be higher than the priority of a job under execution, and the job whose swap priority is high is selected as a job to be swapped out, and memory quantity being the target of swap-out is calculated for each of the nodes of the job to be swapped out based on the required memory quantity of the job whose emergency is designated, and the job to be swapped out is partially swapped out based on the memory quantity designated for each node. Thus, only the memory quantity which is absolutely necessary for execution by the job whose emergency is designated can be selected as the object of swap-out. Also, the partially swapped-out job is re-swapped out with high priority, so that the memory resources can effectively be utilized. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a job swap, and in particular, in a shared memory type SMP (Symmetric MultiProcessor) computer system, a control node (hereinafter referred to as a master node) according to a required memory amount of a job whose urgency is designated and a memory amount of a job to be swapped out. The present invention relates to a job swap technology for controlling a job from.
[0002]
[Prior art]
With an increase in the amount of data processed by the computer system, a multiprocessor computer system that enables various types of parallel processing has been used to speed up the processing. In particular, jobs that perform long-term weather forecasting, collision analysis, computational science, etc. require a huge amount of calculation, and the memory area in which the program itself is loaded is very large. There is a demand for a multiprocessor computer system having a large-capacity virtual memory space for expanding the calculation results and the like. In such a computer system, when an emergency job is input or a forced swapout instruction is issued to execute a job that has been waiting for a long time, the job to be swapped out is selected from the currently running jobs. A job swap function is provided for temporarily releasing resources such as a CPU and a memory, and for activating an urgent job or a job that has been in a long wait state.
[0003]
However, if jobs requiring a large amount of virtual memory space, such as long-term weather forecasts, collision analysis, and computational science, are subject to swap-out, the time required for memory swap-out becomes longer, and the following Techniques for reducing overhead have been proposed.
For example, Japanese Unexamined Patent Publication No. Hei 5-88984 refers to a management book for managing jobs at the time of swapping out a job, and if it is found that the page exists in the main storage device and is in an active state, the swap-out is performed. The address of the target page is registered in the directory, and if it is found that the page exists in the main storage device but is not active, the priority of the page out of the management list is increased, and the page is registered in the directory. By inadvertently transferring pages from main storage to auxiliary storage, inactive pages that are present in main storage at the time of job swap-out but have not been accessed recently are excluded from swapping. The page-out priority is raised and left in the main storage device, and when a page-out request occurs due to a memory request from another job, an LRU (Leas) Recently Used) control swapping control techniques in a virtual storage system which is adapted to preferentially paged out by the page replacement algorithm of the main memory management program of the virtual storage system is disclosed by.
[0004]
Japanese Patent Application Laid-Open No. 2000-200199 discloses a swap-out control unit that creates and manages a swap management table corresponding to a task memory area reservation request and reserves a memory area in response to the memory area reservation request. When it is necessary to perform swap-out on a page-by-page basis or on a task-by-task basis when necessary, control and execute the swap-out (based on the information stored in the swap management table). Setting), the swap-out target limiting means performs a swap-out of “pages that have not been updated and copied” based on the swap management table, when batch swap-out is performed for each task. By controlling not to do this, swap out by page unit and task unit By combining a swapping-out, technology that enables efficient memory management in a virtual memory system is disclosed.
[0005]
Japanese Patent Application Laid-Open No. H11-24950 discloses a PE (processor element) having an external storage device, a PE having no external storage device, a PE having no external storage device, and a PE for executing parallel jobs. In response to the execution instruction, if the resources on the PEs to be executed by the parallel job are insufficient and a low-priority job is being executed, all the PEs that are executing the low-priority job are executed. A process for notifying an instruction to temporarily suspend execution of resources and swap resources to different external storage devices, and a process for temporarily stopping execution of a designated job on each of the notified PEs and instructing the external storage device to execute the job. A running parallel job to be swapped so that a higher priority parallel job can be executed faster by configuring the device to have a process of writing resources to the device. Efficiency satisfactorily write fast technique is disclosed in a plurality of different external storage device on the respective resources.
[0006]
Japanese Patent Application Laid-Open No. 8-16410 discloses, as a related technique, unused PEs are generated in a class that executes a job in which the number of used PEs is less than a limited number and in a class in which a job to be executed has not yet arrived. However, since it is not possible to allocate unused PEs to other classes of jobs in order to reliably allocate PEs to subsequent jobs and jobs arriving later, multiple job queues are used for the purpose of effective use of PEs. A technique for managing a job waiting to be executed and a job being executed, and setting a belonging queue (a queue in which a job is input / executed) and a priority in the queue according to a system resource request amount, a waiting time, and the like of each job. ing.
[0007]
With this technology, the assigned queue and the priority in the queue are dynamically reviewed, jobs are scheduled according to these, and the non-priority execution jobs that are being executed are subject to swapout, and the processor elements that have been swapped out are replaced. Allocated to start a priority execution job waiting to be executed.
[0008]
[Patent Document 1]
JP-A-5-88984 (pages 2-3)
[0009]
[Patent Document 2]
Japanese Patent Application Laid-Open No. 2000-200199 (pages 2-4)
[0010]
[Patent Document 3]
JP-A-11-24950 (pages 2-3)
[0011]
[Patent Document 4]
JP-A-8-16410 (pages 4-5 and 10)
[0012]
[Problems to be solved by the invention]
According to the technique disclosed in Japanese Patent Application Laid-Open No. 5-88984, only the active memory is swapped out at the time of swapping out the memory, and the inactive memory is left in the main memory by increasing the priority of page out. In addition, when a page-out request is generated by a memory request from another job, the overhead of the swapping process is reduced by coping with the page-out request. However, depending on the progress of the job to be swapped out, the ratio of the inactive memory to the memory of the job to be swapped out is not always large, and the amount of memory to be swapped out is the amount of memory required by the urgency designated job. This is often the case.
[0013]
According to the technology disclosed in Japanese Patent Application Laid-Open No. 2000-20019, when swap-out is necessary, swap-out which has been performed only in units of pages in the past is swapped in units of pages. It is determined whether or not to perform swap-out, and the swap-out is controlled based on the determination, but it does not reduce the overhead of the swapping process in consideration of the amount of memory to be swapped out. For this reason, a memory larger than the amount of memory required by the job with the specified urgency is subject to swap-out.
[0014]
The technique disclosed in Japanese Patent Application Laid-Open No. H11-24950 is a technique for efficiently and quickly writing resources to be swapped out to a plurality of external storage devices, and the technique disclosed in Japanese Patent Application Laid-Open No. 2000-200019. Similarly to the above, it does not reduce the overhead of the swapping process in consideration of the amount of memory to be swapped out. For this reason, a memory larger than the amount of memory required by the job with the specified urgency is subject to swap-out.
[0015]
In addition, when the technology disclosed in Japanese Patent Application Laid-Open No. 8-16410 is used, the memory is swapped for each PE (each node) among a plurality of PEs. However, the swap-out target is not taken into account in consideration of the amount of memory to be swapped out, and a memory larger than the amount of memory required by the job specified in the urgency level is to be swapped out.
[0016]
Hereinafter, the problems in the case of using the technology disclosed in Japanese Patent Application Laid-Open No. 8-16410 will be specifically described with reference to FIGS. In FIG. 14, job 0 of swap priority 20 uses nodes 0 to 2 and occupies 20 MB of memory in each node, job 1 of swap priority 40 uses nodes 0 and 1 and each node Occupies 15 MB of memory, the free memory of node 0 is 25 MB, and the free memory of node 1 is 25 MB. Here, the swap priority indicates a priority for preferentially using the resources of each note, and the higher the numerical value, the higher the priority. Here, consider an example in which nodes 0 to 2 of the swap priority 30 designated by the degree of urgency are used, and a job requesting a memory of 30 MB in each node is input.
[0017]
Since the submitted job with the specified urgency has a swap priority of 30, the job 0 having a lower swap priority than that is selected as a swap-out target. Since the amount of memory to be swapped out in each node is in the unit of the node of the job being executed, it is 20 MB for node 0, 20 MB for node 1, and 0 MB for node 2. However, since the amount of memory required by the job specified in the urgency level is 30 MB, the amount of memory to be swapped out of the nodes 0 and 1 has 25 MB of free memory, so that it is sufficient to swap out 5 MB each to 20 MB. More memory is swapped out than necessary.
[0018]
Next, FIG. 15 shows a situation in which job 0 has been swapped out in order to execute the above-mentioned emergency job with respect to FIG. If the total memory size of the swapped-out job is smaller than the required memory size of the swapped-out job, the memory that has not been swapped out (the portion of F1 in the node 2) remains on the computer system. Since this memory cannot be used by other jobs, it is a cause of hindering the effective use of resources, and the prior art has not solved this.
[0019]
The present invention has been proposed in view of the above circumstances, and has as its object to reduce overhead required for job swapping and to effectively use memory resources.
[0020]
[Means for Solving the Problems]
FIG. 1 shows an overall configuration diagram of the first embodiment. The job management program 11 of the present invention is a program to be executed on a computer as the job management apparatus 1, and by partially changing the swap priority of a job that has been swapped out to be lower than that of the running job, the job management program 11 is partially swapped. Priority changing means 111 for increasing the order in which out-printed jobs are swapped out of running jobs, jobs partially swapped out in the order of lower swap priority, and jobs to be swapped out of running jobs , A memory amount calculating unit 113 for calculating a swap-out target memory amount for each node of the swap-out target job based on a required memory amount of an execution waiting job whose urgency is specified, The swap-out pair based on the amount of memory specified for each By having the swap means 114 for partially swapping out a job, only the minimum amount of memory required for the execution of a job designated for urgency can be set as a swap-out target. Overhead can be minimized. Also, by partially swapping out jobs that have been partially swapped out again, it is possible to effectively use memory resources.
[0021]
The job management program 11 is provided with a memory monitoring means 115 for monitoring the free memory of each node, and when it is detected that the free memory amount of each node becomes smaller than a predetermined value, the partial swap-out is performed. By calling the swap means 114 based on a predetermined rule and swapping out the memory of the job which has not been swapped out, it is possible to further effectively use the memory resources.
[0022]
Further, by changing the swap priority of the partially swapped-out job to be higher than that of the job whose entire memory has been swapped out, the priority changing means 111 allows the partially swapped out job to be changed. All the memories have a function of increasing the order of swap-in from the jobs which have been swapped out, and the job selecting means 112 has a function of selecting a job to be swapped in from the jobs which have been swapped out in the order of higher swap priority. By providing the swap means 114 with a function of swapping in the swap-in target job based on the amount of memory specified for each node, the time when an unusable memory occurs can be reduced, and the memory resources can be further reduced. Effective utilization can be achieved.
[0023]
BEST MODE FOR CARRYING OUT THE INVENTION
FIG. 1 is an overall configuration diagram of Embodiment 1 of the present invention. The job management program 11 of the present invention is a program executed on a computer acting as the job management device 1. The priority changing unit 111 changes the swap priority of the partially swapped-out job to be lower than that of the running job in order to increase the order of the partially swapped-out job and the running job. To reduce the swap priority of partially swapped-out jobs to all memory, partially swapped-out jobs have higher priority than all swapped-out jobs Change higher than the job that was swapped out.
[0024]
The job selection unit 112 selects a job to be swapped out in the order of the highest swap priority from the partially swapped-out job and the running job, or performs the swapping from the job swapped out in the order of the lowest swap priority. The in-target job is selected, and the memory amount calculation unit 113 calculates the swap-out target memory amount for each node of the swap-out target job.
[0025]
The swap unit 114 partially swaps in / out the job to be swapped in / out based on the memory amount designated for each node. The memory amount monitoring unit 115 monitors the free memory amount of each node, and If the memory amount satisfies a predetermined condition, the memory that has not been swapped out is swapped out, or the memory that has not been swapped out is swapped out when the swap-in job cannot perform the swap-in due to insufficient memory.
[0026]
The job memory management table 12 exists on the memory of the master node 1 and manages the memory for job swapping for each job. The node memory management table 13 is a table for managing free memory for each node. The job execution device (job execution node node 0) 2 and the job execution device (job execution node node 1) 3 are computers that execute a user's job, and the job management device (master node) is an inter-node connection device. Are connected to each other.
[0027]
FIG. 2 shows the data format of an example of the job memory management table. This table is managed for each process in a job. The contents to be managed are as follows. What is managed for each job includes a swap priority and a swap status area. The swap priority sets the priority of using the resources (memory, CPU) of the node. If the swap priority of the urgent job is higher than the swap priority of the running job, the running job is swapped out. It is also used for the priority of jobs to be swapped in. In the swap state area, a flag is set to determine whether all the memories of the job to be swapped out are swapped out or partially swapped out.
[0028]
Items managed for each process include a node number, a required memory amount, a memory amount to be swapped out this time, and a total memory amount which has been swapped out. The node number is a node number where the process executes. The required memory amount is a memory amount required for each process, and is the same between processes. The amount of memory to be swapped out this time sets the amount of memory for the job to be swapped out. The amount of memory swapped out differs for each process. The swapped out total memory size sets the memory size of the job that has been swapped out.
[0029]
The amount of memory for the job to be swapped out is calculated in page units from the amount of memory required by each process of the job with the specified urgency, and the minimum swapout required to execute the job with the specified urgency is calculated. Set the amount of memory in page units for each process.
FIG. 3 shows a data format of an example of a node memory management table for managing free memory for each node. The node number is set to the number for identifying the node, the memory mounting amount is set to the memory amount mounted on the node, and the free memory amount is set to the value of the current free memory.
[0030]
FIG. 4 is a flowchart showing a flow of the partial swap-out process according to the first embodiment. When the urgency level is specified for the job waiting to be executed (S401), a job list for swap-out is created based on the swap-out state and the swap priority (S402). Details will be described in S408 to S412.
Next, the amount of memory to be swapped out from the top job in the swap target job list is calculated for each node based on the amount of memory requested by the urgent degree designated job (S403), and the urgent degree designated job operates. Is set as the memory amount to be swapped out (S404). Details will be described in S413 to S416.
[0031]
Next, it is determined whether or not the amount of memory that can be executed by the urgency-designated job has been reached (S406). If not, the address of the job memory management table for the next job in the swap-out target job list is set to the job. The memory management table pointer is set (S405), and the process returns to S403. If it has reached, the selected job is swapped out (S407). Since the swap-out process itself is a conventional technique, details of the process are omitted.
[0032]
The swap-out target job list is created for running jobs whose swap priority is lower than the job with the specified urgency level and for jobs that have been partially swapped out. It is determined whether or not the flag in the swap status area is a job in which partial swap-out is being performed (S408). If the job is being partially swapped out, the swap priority is changed to a value smaller than the job being executed in order to increase the swap-out priority (S410).
[0033]
If not, it is determined whether or not the swap priority of the running job is smaller than the job of which the urgency level is specified (S411). If it is smaller, the swap priority is raised compared to the swap priority of other running jobs, and if smaller, the swap priority is increased (S412), and the address of the next job memory management table is changed to the job memory to check the next job. It is set to the management table pointer (S409), and the process returns to S408. If not, the address of the next job memory management table is set in the job memory management table pointer to check the next job (S409), and the process returns to S408. When all the jobs in the running job queue and the jobs in the swout-out job queue have been checked, the address of the job memory management table of the first job in the job list to be swapped out is set in the job memory management table pointer, and the process proceeds to S403.
[0034]
To calculate the amount of memory to be swapped out for each node based on the amount of memory requested by the urgent level specified job from the top of the list of jobs to be swapped out, Is determined by referring to the amount of free memory in the node memory management table (S413). If not, the required amount of memory is calculated for each node (S415), the value is set to "the amount of memory to be swapped out this time" in the job memory management table (S416), and the next node is checked. To do so, the address of the next node memory management table is set in the node memory management table pointer (S414), and the process returns to S413.
[0035]
If it is sufficient, the address of the next node memory management table is set in the node memory management table pointer to check the next node (S414), and the process returns to S413. The amount of memory to be swapped out is a value obtained by subtracting the amount of free memory of the node from the amount of memory required for the urgency designated job. When the check of all the nodes is completed, the process proceeds to S406.
[0036]
A specific example of the above-described swap processing will be described based on the example of the job memory management table before the swap-out shown in FIG. 5 and the example of the job memory management table after the swap-out shown in FIG.
At present, it is assumed that there is one running job with a swap priority of 20, the free memory of node 0 is 0 MB, and the free memory of node 1 is 48 MB. Here, an example of the job memory management table before the running job is swapped out when the urgent designation job whose required memory amount per process is 32 MB, the number of processes is 3, and the swap priority is 30 is input. This is shown in FIG. FIG. 6 shows an example of the job memory management table after the running job is swapped out. Since the node number of the process 0 of the job with the specified urgency is 0, the required memory is 32 MB, and the free memory of the node 0 is 0, the memory amount to be swapped out this time is 32 MB. The node numbers of both processes 1 and 2 of the urgency designated job are 1, the required memory amounts are 32 MB, and the free memory amount of node 1 is 48 MB. Therefore, the shortage memory of processes 1 and 2 is 16 MB in total. Become. It is necessary to swap out 16 MB from process 1, but not in process 2.
[0037]
FIG. 6 shows an example of the job memory management table after the running job has been swapped out. As a result of the swap-out, the total memory amount of the process 0 that has been swapped out is 32 MB, that of the process 1 is 16 MB, and that of the process 2 is 0 MB. After the swap-out, the swap status area of the job is changed to a memory partial swap.
[0038]
FIG. 7 is a flowchart showing a flow of the automatic swap-out processing according to the first embodiment. The free memory amount of each node is periodically monitored by, for example, a timer monitor, and when the free memory amount becomes smaller than a predetermined amount, the memory that has not been swapped out is automatically swapped out (S701) to use the memory resources. Try to use it. Whether or not the free memory is smaller than the predetermined value is determined, for example, based on whether or not the free memory amount of the node is less than 1/10 of the memory amount mounted on the node (S705).
[0039]
The amount of memory that is automatically swapped out is swapped out, for example, by setting the memory that has not been swapped out in the target node until the free memory amount becomes half (S707). In order to perform this for all nodes, the address of the next node memory management table is set in the node memory management table pointer (S706), and the process returns to S705. When all the nodes have been checked, the process proceeds to S702.
[0040]
Next, swap-in jobs are selected in descending order of swap priority (S702). If the selected swap-in job cannot be swapped in due to lack of memory, the memory that has not been swapped out is automatically swapped out (S703). Details will be described in S708 to S712. When the swap-in memory is secured, the selected job is swapped in (S704). Since the swap-in processing itself is a conventional technique, details of the processing are omitted.
[0041]
The details of S703 will be described. It is determined whether or not the required memory size of the swap-in job is free in the node by referring to the node memory management table (S708). If it is free, the address of the next node memory management table is set in the node memory management table pointer to check the next node (S709), and the process returns to S708. When all the nodes have been checked, the process proceeds to S704.
[0042]
If not free, the amount of memory required for swap-in is calculated for each node by subtracting the amount of free memory of the node from the amount of required memory (S710), and a job with a large amount of memory that has not been swapped out is selected (S710). S711), the value is set to “the amount of memory to be swapped out this time” in the job memory management table, swap-out processing is performed (S712), and the address of the next node memory management table is set to check the next node. The node memory management table pointer is set (S709), and the process returns to S708. Since the swap-out process itself is a conventional technique, the details of the process are omitted.
[0043]
FIG. 8 shows an example of the node memory management table before the automatic swap-out by monitoring the free memory of the node. Since the free memory of node 0 is less than one-tenth of the mounted memory, automatic swap-out is performed until the amount of non-swapped-out memory of node 0 becomes one-half or more of the mounted memory. Done. In this example, as shown by the following equation, a memory amount of 108 MB is subject to automatic swap-out.
(256 MB ÷ 2) −20 MB = 108 MB
Since the free memory amount of the node 1 is larger than one tenth of the mounted memory amount, the node 1 is not subjected to the automatic swap-out.
[0044]
FIG. 9 shows an example of the node memory management table after the automatic swap-out by monitoring the free memory of the node. Due to the automatic swap-out, the free memory capacity of the node 0 becomes 128 MB.
FIG. 10 shows an example of the job memory management table before the automatic swap based on the required memory amount of the swap-in job.
[0045]
When the free memory capacity of the node 0 is 64 and the free memory capacity of the node 1 is 56 MB, the required memory capacity per process of the swap-in job is 32 MB, and the number of processes is 3 (32 × 3 = 96 MB per job). In some cases, since the memory capacity of the swap-in job is insufficient, the amount of memory to be swapped out for each process among the nodes in the swap memory capacity is calculated.
[0046]
a) The amount of memory to be swapped out this time for process 0 using node 0 is
Requested memory amount (32 MB) −free memory amount of node 0 (64 MB) = memory amount to be swapped out (0 MB)
In this case, since the amount of memory to be swapped out is 0, no swap-out is performed.
[0047]
b) The amount of memory to be swapped out this time for the process using node 1 is
Requested memory amount (32 MB × 2) −free memory amount of node 1 (56 MB) = 8 MB
And the amount of memory to be swapped out this time in process 1 is 8 MB. Also,
The amount of memory to be swapped out this time in process 2 is 0 MB.
FIG. 11 shows an example of the job memory management table after the automatic swap based on the required memory amount of the swap-in job. In the process 1, 8 MB is added to the total memory amount of the swapped-out memory, resulting in 24 MB. Since the amount of memory is being partially swapped out for the entire job, the swap state area remains in the memory partial swapping.
[0048]
FIG. 12 is a flowchart showing the flow of the partial swap-in process according to the first embodiment. When resources are released to the computer system due to the termination of the job being executed or the like, the swapped-out jobs can be swapped in (S121). Therefore, the jobs being swapped out are arranged in the order of the swap priority (S122). ).
[0049]
Next, for the job having the highest swap priority, it is determined whether there is another job having the same swap priority (S123). If not, the job having the highest swap priority is swapped in (S125). ). If there is another job having the same swap priority, a job to be swapped in is selected according to the amount of memory that has not been swapped out as shown in S127 to S1211 (S124). Since the swap-in processing itself is a conventional technique, details of the processing are omitted.
[0050]
In order to select a job to be swapped in according to the amount of memory that has not been swapped out, the job memory management table is first referred to, and it is determined whether or not the flag in the swap status area is in the process of swapping memory (S127). If not, the address of the job memory management table for the next swap-out job is set in the job memory management table pointer (S126), and the process returns to S127.
[0051]
If the memory is being partially swapped, the swap priority is changed to be higher than that of the job in which all memory is swapped in order to raise the swap-in priority of the flag in the swap state area over the job in which all memory is swapped (S128). . Next, the amount of memory that has not been swapped out is calculated by subtracting the swapped out total memory from the required memory amount of the job (S129). Next, the calculated amount of memory that has not been swapped out is compared with other swap-out jobs (S1210). If the calculated amount is smaller, the address of the job memory management table of the next swap-out job is set in the job memory management table pointer. (S126), and returns to S127 (S126). If it is larger, the swap priority is changed to a value larger than the current value so as to further increase the swap-in priority (S1211).
[0052]
When the swap priority is changed in this way, as shown in FIG. 13, the larger the amount of memory that has been partially swapped out and has not been swapped out, the higher the swap-in order, and the more efficient the use of memory resources. It becomes possible to plan.
(Supplementary Note 1) A job swap method for a shared memory type SMP computer system, comprising: a step of changing the swap priority of a partially swapped out job based on a predetermined rule; Selecting a job to be swapped out from the jobs that have been swapped out, and a job to be swapped out. A job swap method, comprising: calculating a memory amount; and partially swapping out the swap-out target job based on a memory amount designated for each node.
[0053]
(Supplementary note 2) The job swap according to Supplementary note 1, wherein when the free memory capacity of each node satisfies a predetermined condition, the non-swapped-out memory of the partially swapped-out job is swapped out based on a predetermined rule. Method.
(Supplementary Note 3) When it is detected that the memory amount required by the swap-in job is larger than the free memory amount, the non-swapped-out memory of the partially swapped-out job is swapped out based on a predetermined rule. 1. The job swap method according to 1.
[0054]
(Supplementary Note 4) A job swapping method for a shared memory type SMP computer system, wherein a step of changing the swap priority of a partially swapped out job based on a predetermined rule and a step of swapping out based on the swap priority Selecting a swap-in target job from the selected jobs; and swapping the swap-in target job based on the amount of memory specified for each node.
And a job swap method.
[0055]
(Supplementary Note 5) Priority changing means for changing the swap priority of a partially swapped-out job based on a predetermined rule, a job partially swapped out based on a swap priority, and a running job A job selecting means for selecting a job to be swapped out from, and a memory amount calculating means for calculating a memory amount to be swapped out for each node of the job to be swapped out based on a required memory amount of a job waiting to be executed designated by urgency And a swap unit for partially swapping out the swap-out target job based on a memory amount designated for each node.
[0056]
(Supplementary Note 6) Priority changing means for changing the swap priority of a partially swapped-out job based on a predetermined rule, a job partially swapped out based on the swap priority, and a running job A job selecting means for selecting a job to be swapped out from, and a memory amount calculating means for calculating a memory amount to be swapped out for each node of the job to be swapped out based on a required memory amount of a job waiting to be executed designated by urgency And a job management program for causing a computer to function as swap means for partially swapping out the swap-out target job based on a memory amount designated for each node.
[0057]
(Supplementary Note 7) A step of changing a swap priority of a partially swapped-out job based on a predetermined rule, and a priority change of selecting a job to be swapped in from the jobs swapped out based on the swap priority And a swap unit for swapping in the swap-in target job based on a memory amount designated for each node.
[0058]
(Supplementary Note 8) A step of changing a swap priority of a partially swapped-out job based on a predetermined rule, and a priority change of selecting a swap-in target job from the swapped-out jobs based on the swap priority Means and a job management program for causing a computer to function as a swap means for swapping in the swap-in target job based on a memory amount designated for each node.
[0059]
【The invention's effect】
According to the present invention, it is possible to reduce the overhead required for job swap and to effectively use memory resources in a shared memory type SMP computer system in which a single node or a plurality of nodes are connected by a network.
[Brief description of the drawings]
FIG. 1 is an overall configuration diagram of a first embodiment.
FIG. 2 shows an example of a job memory management table.
FIG. 3 is an example of a node memory management table;
FIG. 4 is a partial swap-out process according to the first embodiment;
FIG. 5 shows an example of a job memory management table before swap-out.
FIG. 6 shows a job memory management table after swap-out.
FIG. 7 is an automatic swap-out process according to the first embodiment;
FIG. 8 shows an example of a node memory management table before automatic swap-out.
FIG. 9 is an example of a node memory management table after automatic swap-out
FIG. 10 shows an example of a job memory management table before automatic swap-out.
FIG. 11 shows an example of a job memory management table after automatic swap-out.
FIG. 12 is a partial swap-in process according to the first embodiment;
FIG. 13 shows an example of a swap-in target job list.
FIG. 14 shows an example of overhead reduction at the time of swap-out according to the related art.
FIG. 15 shows an example in which resources are not effectively used in the conventional technology.
[Explanation of symbols]
1 Job management device (master node)
2 Job execution device (job execution node node 0)
3 Job execution device (Job execution node Node 1)
4 Node connection device
11 Job management program
12 Job memory management table
13 Node memory management table
111 Priority change means
112 Job selection means
113 Memory amount calculation means
114 Swap Means
115 Memory amount monitoring means

Claims (5)

A job swap method for a shared memory type SMP computer system,
Changing the swap priority of the partially swapped-out job based on a predetermined rule;
Selecting jobs to be swapped out of partially swapped out jobs and running jobs based on swap priority;
Calculating a swap-out target memory amount for each node of the swap-out target job based on the requested memory amount of the execution waiting job specified in the urgency degree;
Partially swapping out the job to be swapped out based on the amount of memory specified for each node. 2. The job swapping method according to claim 1, wherein when the free memory capacity of each node satisfies a predetermined condition, the non-swapped-out memory of the partially swapped-out job is swapped out based on a predetermined rule. A job swap method for a shared memory type SMP computer system,
Changing the swap priority of the partially swapped-out job based on a predetermined rule;
Selecting a job to be swapped in from jobs that have been swapped out based on swap priority;
Swapping the swap-in target job based on the amount of memory specified for each node. Priority changing means for changing the swap priority of a partially swapped-out job based on a predetermined rule;
A job selection unit for selecting a job that has been partially swapped out based on the swap priority and a job to be swapped out of running jobs,
A memory amount calculation unit configured to calculate a swap-out target memory amount for each node of the swap-out target job based on the requested memory amount of the execution waiting job specified in the urgency degree;
And a swap unit for partially swapping out the job to be swapped out based on a memory amount designated for each node. Priority changing means for changing the swap priority of a partially swapped-out job based on a predetermined rule;
A job selection unit for selecting a job that has been partially swapped out based on the swap priority and a job to be swapped out of running jobs,
A memory amount calculation unit configured to calculate a swap-out target memory amount for each node of the swap-out target job based on the requested memory amount of the execution waiting job specified in the urgency degree;
A job management program for causing a computer to function as swap means for partially swapping out the swap-out target job based on a memory amount designated for each node.

Images