JPH0883257A - Process execution optimization method for parallel computer systems - Google Patents

Abstract

(57) [Summary] [Purpose] Dynamically move the process of a processor with a high utilization rate to a processor with a low utilization rate in the middle of execution to efficiently process jobs. [Configuration] A management processor 11 and a plurality of processing processors 12, 13, and 14 are connected by a network 15. The memory 16 of the management processor 11 includes a priority management table 20 that defines priorities when moving processes, an execution process of each processing processor, and an execution process management table 30 that manages an execution utilization rate. According to the execution process management table 30, the management processor 11 determines, when there is a processing processor with a high usage rate and a processing processor with a low usage rate, from among a plurality of processes being executed by the processor with a high usage rate. The process with the lowest priority is selected by referring to the priority management table 20, and the process is moved to the processor with a low utilization rate.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a parallel computer system in which a plurality of processors are connected to a high-speed network and execute a plurality of processes for a plurality of jobs at the same time. Regarding the optimization method.

[0002]

2. Description of the Related Art Generally, when a task is executed in a parallel computer system, a processor having a high utilization rate (a load is concentrated) and a processor having a low utilization rate (a light load or no load) occur. In such a case, a conventional computer system of this type can automatically perform a process by moving a process running on a processor with insufficient performance to another processor with sufficient performance without operator operation. There wasn't.

[0003]

In the prior art,
When changing the distribution of processes executed by each processor in a parallel computer system, it requires an operator's operation one by one, so if there is a processor with a high utilization rate and a processor with a low utilization rate, the utilization There has been a problem that a part of the process running on the high processor cannot be efficiently moved by dynamically moving to a processor with a low utilization rate.

It is an object of the present invention to efficiently and efficiently process a process of a processor having a high utilization rate to a processor having a low utilization rate in a parallel computer system during execution while automatically and dynamically without an operator's operation. To provide a process execution optimization method.

[0005]

SUMMARY OF THE INVENTION The present invention comprises a plurality of processing processors for processing a plurality of processes in parallel, and a management processor for managing the execution processes of these processing processors. The priority order of the processes processed by each processing processor and the execution utilization rate of each processing processor are retained, and the priority order is automatically selected from the multiple processes being executed by the processing processors with high execution utilization rates. That is, a process with a low execution rate is selected and moved to a processing processor with a low execution utilization rate.

[0006]

[Operation] The management processor receives from each processing processor,
The execution utilization rate at that time is collected at regular intervals or as needed. Then, according to the collected execution utilization rate of each processing processor, the execution of the process having the lowest priority is stopped from the plurality of processes being executed by the processor having a high execution utilization rate (the load is concentrated), As a result of execution up to the stop, the program stop point is transferred together with the process to a processor with a low execution utilization rate (low load). The processor having a low execution utilization rate continues the processing of the transferred process based on the execution result up to the stop and the program stop point. The management processor may be an arbitrary processor among the plurality of processing processors.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a system configuration diagram of an embodiment of the present invention. Processors 11 to 14 are connected to a high speed network 15. In the present embodiment, the processor 11 is a management processor, the processors 12 to 14 are process execution processors, and the memory 16 of the management processor 11 has a priority for managing the priority of the process executed by each processing processor. A degree management table 20 and an execution process management table 30 for managing the execution process of each processing processor, the prescribed usage rate, and the execution usage rate are prepared. ~
Indicates a currently executing process on the processing processors 12 to 14.

It should be noted that all the processors 11 to 14 may be processing processors, and any of the processors may also function as a management processor. This has the advantage that if the management processor fails, its function can be taken over by another processor.

FIG. 2 shows the memory 1 of the management processor 11.
6 shows an example of the configuration of the priority management table 20 shown in FIG.
Reference numeral 23 is a process ID corresponding to each execution work, and 24 is a priority order of processes. In the case of this example, the work priority is A>B> C, and the process priority is>,
>,>. The criteria for prioritizing may be the importance of the business or process or the urgency,
It may be variable depending on the system used.

FIG. 3 shows an example of the configuration of the execution process management table 30 also on the memory 16 of the management processor 11, where 31 is the processor number and 32 is the currently executing process (execution process) in each processor. ), 33
Is a specified utilization rate (MAX) that defines the upper limit of the load that is a guideline for determining the processor that reduces the execution process for each processor, and 34 is a specified usage that defines the lower limit of the load that is a guideline that determines the processor to which the execution process is moved. Rate (M
IN), 35 is the actual load utilization rate (execution utilization rate) of each processor. The execution utilization rate is defined by, for example, the number of memory accesses per certain time (access frequency).

In this embodiment, processor numbers 1 to 1 in FIG.
3 corresponds to the processing processors 12, 13, and 14 of FIG. From FIG. 3, the processor 12 is a process, the processor 13 is a process, and the processor 14 is
Executes processes and their execution utilization rates 35 are 90%, 10% and 30%, respectively. On the other hand, the specified utilization rates 33 and 34 are set to 80% and 20% uniformly for the processors 12, 13, and 14 here.

The management processor 11 refers to the execution process management table 30, and there is a processing processor having an execution utilization rate of 80% or more (the load is concentrated), and further, an execution utilization rate of 20% or less (load). If there is a processing processor with a low priority, the process with the lowest priority is referred to by referring to the priority management table 20 from the plurality of processes being processed by the processing processor with an execution utilization rate of 80% or more. Is selected, and the processing is moved to a processing processor having an execution utilization rate of 20% or less. In FIGS. 2 and 3, the processing of the process of the processing processor 12 having the processor number 1 is the processing processor 1 of the processor number 2.
Moved to 3.

FIG. 4 is a flow chart showing the flow of processing from distribution of the process to each processing processor to movement of the process according to the execution utilization rate of the processor, and FIG. 5 at that time is exchanged between the management processor and the processing processor. It is a figure which shows the data format performed. Hereinafter, FIGS. 4 and 5
A process for automatically moving a process being executed by a processor having a high execution utilization rate to a processor having a low execution utilization rate will be described with reference to FIG.

The management processor 11 scales the execution work and distributes the process to be executed to each processing processor 12, 13, 14 (step 401). Figure 5
501 indicates a data format at the time of distributing this execution process. Execution process distribution is per processor,
Although it may be broadcast, the execution process IDs of other processors are also distributed to each processor due to the necessity of communication between processes. Also, the management processor 11 uses the memory 1 when distributing the execution process.
6 sets the priority for each business and each process in the priority management table 20 and stores the process executed by each management processor 12, 13, 14 in the execution process management table 30. The specified utilization rates 33 and 34 of the execution process management table 30 may be set in advance, or may be set in a pair with the execution process in the step 401. The execution utilization rate 35 of each processor is all zero in the initial state.

Each processing processor 12, 13, 14 is
The processing of the process distributed from the management processor 11 is started (step 402). In this embodiment, the processor 12 having the processor number 1 starts processing, the processor 13 having the processor number 2 starts processing, and the processor 14 having the processor number 3 starts processing.

Each processing processor 12, 13, 14 is
A storage device for storing various data is provided, and the process is processed by accessing the storage device.
If necessary, it communicates with the process of another processor to access the storage device of the other processor. Here, the access frequency to this storage device is used as a measure of the execution utilization rate (degree of load concentration) of the processor. When the processing of each process is started, each of the processing processors 12, 13, 14 counts the number of times of access to the storage device per unit time, and holds the latest value as the execution utilization rate. It should be noted that the execution utilization rate may be based on the number of instruction processings per fixed time.

The management processor 11 collects the above-mentioned execution utilization rates of the processing processors 12, 13, and 14 and updates the execution utilization rate 35 of the execution process management table 30 (step 403). The collection of execution utilization is
The management processor 11 issues a request to each of the processing processors 12, 13, 14 at regular time intervals, and in response, each of the processing processors 12, 13, 14 sends the execution utilization rate at that point to the management processor 11. I will do so by contacting me. Reference numerals 502 to 504 in FIG.
Reference numerals 2, 13 and 14 show data formats when the execution utilization rate is notified to the management processor 11.

When the execution utilization rates of the processing processors 12, 13, 14 are collected, the management processor 11 determines the prescribed utilization rate (MAX) 3 in the execution process management table 30.
3 is compared with the effective utilization rate 35 to check whether there is a processor having an effective utilization rate of 80% or more (step 404). Here, if there is no processor having an execution utilization rate of 80% or more, it is considered that each of the processing processors 12, 13, and 14 has no load concentration, and in this case, nothing is done. That is, the processing processors 12, 13 and 14 continue the processing as they are.

On the other hand, if there is a processor whose execution utilization rate is 80% or more, it is considered that the load is concentrated on the processor. In this case, the management processor 11 compares the specified usage rate (MIN) 34 and the execution usage rate 35 in the execution process management table 30 to check whether there is a processor with an execution usage rate of 20% or less (step 405). . That is, a processor with a light load is searched for before moving processes of a processor with a heavy load. Also here, if there is no processor whose execution utilization rate is 20% or less (the load is light), nothing is done. In this case, each processing processor 1
Since the processing of Nos. 2, 13 and 14 continues as it is, the processor having the concentrated load will continue the process processing in the state of the concentrated load. That is, if there is a processor with a utilization rate of 80% or more (load concentration) but no processor with a utilization rate of 20% or less (light load), even if the process is moved, the execution utilization rate of the migration destination processor is 80%.
Since there is a high possibility that it will be more than%, and the processing efficiency will eventually deteriorate, the processing will continue without moving.

When there is a processor having an execution utilization rate of 80% or more (load concentration) and a processor having an execution utilization rate of 20% or less (light load), the management processor 11 first determines the execution utilization rate. Among the execution processes of 80% or more of the processors, the process with the lowest priority is selected by referring to the priority management table 20, and the processors having the execution utilization ratio of 80% or more are caused to stop the processing of the process. The program stop points are collected as a result of execution up to the point of stopping (step 406). Next, the management processor 11 sets the process contents in which the processing is stopped by the processor having the above-mentioned execution utilization rate of 80% or more, the execution result up to that point, and the program stop point for the processor with the execution utilization rate of 20% or less. Is transferred (step 407).

In FIGS. 2 and 3, the execution utilization rate of the processing processor 12 having the processor number 1 is 90%, and the execution utilization rate of the processing processor 13 having the processor number 2 is 10%.
Further, the priority of the process being executed by the processing processor 12 is>. Therefore, the management processor 11 causes the processing processor 12 having an execution utilization rate of 90% to stop the execution of the process, collects the execution result up to the point of the stop, the program stop point, and completes the processing. Then, the process and the data up to the time when the process is stopped (execution result, program stop point) are transferred to the processing processor 13 having an execution utilization rate of 10%. Reference numeral 505 in FIG. 5 indicates a data format for notifying the processing processor 12 of the process stop, 5
Reference numeral 06 denotes a data format for collecting data of the stopped process from the processing processor 12, and 507 denotes a data format for transferring desired data to the processing processor 13 of the process migration destination.

After transferring the process and the desired data to the processor for processing having an execution utilization rate of 20% (light load),
The management processor 11 uses the execution process management table 3
The execution process column 32 of 0 is rewritten, the process move destination processor is notified to each processing processor 12, 13, 14 (address change notification), and the process start instruction is issued to the process move destination processing processor (step). 408). Here, in the execution process management table 30 of FIG. 3, the process is deleted from the execution process of the processor number 1 (processor 12), and the process is added to the execution process of the processor number 2 (processor 13). Further, it notifies each processing processor that the processing processor of the process has been changed to the processor number 2 (processor 13), and instructs the processor 13 of the processor number 2 to start processing of the process. Reference numeral 508 in FIG. 5 is a data format used when the management processor 11 notifies each processing processor 12, 13, 14 of a change (address change) of the processing processor of the process by broadcasting or the like, and 509 is a management processor. Reference numeral 11 shows a data format when the process processor 13 is instructed to start the process.

Upon receiving the process start instruction, the processing processor restarts the process (step 4).
09). Here, the processing processor 13 with the processor number 2 restarts the process from the transferred program stop point.

Generally, when M is the number of high-load processors with an execution utilization rate of 80% or more and N is the number of low-load processors with an execution utilization rate of 20% or less, the high-load processor executes if M ≦ N. The low-load processor may move the target process having a low priority by associating the process source / destination processors in order from the highest utilization factor to the lowest utilization factor. Also,
In the case of M> N, N processors are selected from the M high-load processors in descending order of execution utilization rate, and are selected as N low-load processors in the same manner as in the case of M ≦ N. The target processes may be moved in association with each other.

The processing of steps 404 to 409 in FIG.
It is repeated every time the execution utilization rate of each processing processor 12, 13, 14 is collected. As a result, it is possible to dynamically distribute the processes of a processor with a high execution usage rate (where the load is concentrated) to a processor with a low execution usage rate (the load is low) and continue processing. . At the time of this distribution, the process of the migration target process is temporarily stopped. However, since the process with the lowest priority is selected from the multi-processes that are being executed by the processor with a high execution utilization rate, the process is affected. Can be minimized.

As described above, in the description of the embodiment, the management processor issues a request to the processing processors at regular time intervals to collect the execution usage rates, and in response thereto, each processing processor determines the execution usage rate. However, each processing processor sends the execution utilization rate irregularly (arbitrarily), and the management processor receives it and receives it each time. Alternatively, the execution utilization rate 35 corresponding to the corresponding processor number in the execution process management table may be updated. in this case,
The sequence for issuing a transmission request of the execution utilization rate from the management processor to each processing processor is omitted. In addition,
In this case, the processing of steps 404 to 409 of FIG. 4 may be performed every time the execution utilization rate is collected from a certain processing processor, or may be performed irrespective of that at regular time intervals.

[0028]

According to the process execution optimizing method of the parallel computer system of the present invention, the following effects can be obtained.

(1) Depending on the execution utilization rate of the processing processor, the process under execution automatically and dynamically moves from the processor with a high utilization rate (concentrated load) to the processor with a low utilization rate (low load). By letting
It is possible to use the processor efficiently and process the process efficiently, and it is possible to reduce the processing time of the execution work.

(2) A process to be moved from a processor with a high utilization rate to a processor with a low utilization rate is based on the priority of the process depending on the priority of a predetermined job and the like. Since the process with the lowest priority among the processes is targeted, it is possible to minimize the influence on the business to be executed.

(3) When a management processor moves a process from a process having a high usage rate to a processor having a low usage rate, the management processor notifies the process migration destination processor together with the process to all the processing processors to perform subsequent processing. When inter-process communication becomes necessary, the processing processor can perform inter-process communication without the intervention of the management processor.

(4) By adopting a configuration in which any of the plurality of processing processors also serves as the management processor, when the management processor fails, the management information can be transferred to another processor. It becomes possible to continuously move the system by using the processor as a management processor.

[Brief description of drawings]

FIG. 1 is a block diagram of one embodiment of a parallel computer system according to the present invention.

FIG. 2 is a diagram showing a configuration example of a priority management table on a management processor.

FIG. 3 is a diagram showing a configuration example of an execution process management table on a management processor.

FIG. 4 is a process flow diagram of an embodiment of a process execution optimization method according to the present invention.

FIG. 5 is a diagram showing an example of a data format exchanged between a management processor and a processing processor.

[Explanation of symbols]

 11 Management Processor 12-14 Processing Processor 15 Network 16 Management Processor Memory 20 Priority Management Table 30 Execution Process Management Table

Claims (3)

[Claims] 1. A process execution optimum in a parallel computer system comprising a plurality of processing processors for processing a plurality of processes in parallel, a management processor for managing execution processes of the processing processors, and a network connecting the processors. The management processor, wherein the management processor holds the priority order of the processes processed by each processing processor and the execution utilization rate of each processing processor, and is executed by the processing processor having a high execution utilization rate. A process execution optimization method for a parallel computer system, which comprises selecting a process having a low priority from a plurality of processes and moving the selected process to a processing processor having a low execution utilization rate. 2. The process execution optimization method for a parallel computer system according to claim 1, wherein the management processor communicates a process destination processor to each processing processor. Execution optimization method. 3. The process execution optimizing method for a parallel computer system according to claim 1, wherein any one of the plurality of processing processors also serves as the management processor. System process execution optimization method.