JPH09167142A - Job dispatch device for parallel computer system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently and parallelly execute plural efficient jobs by processing the job while distributing the load by a CPU group for which CPU dedicated to the execution of single job are parallelly connected. SOLUTION: In a job dispatch device for parallel computer system with which the plural jobs are processed while distributing the load by plural CPU, a job managing CPU 1 is provided with an input spool 2 for accepting plural jobs 9-12 and an output spool 3 for accepting the output processing of output job 17. Job execution CPU 13-16 form the CPU group, for which the plural CPU dedicated to the execution of single job are parallelly connected, and a job managing CPU 1 performs the allocation of jobs 9-12 to the job execution CPU 13-16. Besides, each CPU of plural job execution CPU 13-16 forming the CPU group is provided with a busy bit showing whether this CPU executes the job or not.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a job dispatch apparatus for a parallel computer system which executes a plurality of jobs by a plurality of CPUs, and more particularly, a CPU in which a plurality of CPUs dedicated to executing a single job are connected in parallel. The present invention relates to a technique effectively applied to a job dispatching device of a parallel computer system for load-balancing and processing a plurality of jobs by a group.

[0002]

2. Description of the Related Art In a conventional computer system, a plurality of jobs are executed in parallel on one CPU by a method called multitasking. A plurality of jobs are executed in parallel by executing them on one CPU for a certain period of time, and the waiting time of the CPU generated during the execution of the jobs is reduced as much as possible.

In the multi-tasking, a task of executing a specific job of the plurality of jobs for a certain period of time is performed, then control is transferred to another job, and a job of transferring the control is performed for a certain period of time. The processing is repeated until the processing of the plurality of jobs is completed.

Therefore, in the computer system which performs the multitasking, it is necessary for the operating system to perform the task switching process for transferring the control from the task that executes a certain job to the task that executes another job. Since such processing called overhead performed by the operating system consumes processing time separately from the processing for executing the plurality of jobs, the execution of the processing for the plurality of jobs is delayed as the overhead increases.

[0005]

The present inventor has found the following problems as a result of examining the above-mentioned prior art.

That is, when a plurality of jobs are executed by one CPU by multitasking in the conventional computer system, a task is generated by executing a plurality of jobs in parallel on the one CPU in a time-sharing manner. The switching process causes a great overhead in the entire computer system, which causes a great obstacle to the expansion of the computer system as the job processing amount increases.

An object of the present invention is to provide a technique capable of realizing efficient parallel execution of a plurality of jobs.

[0008] The above and other objects and novel features of the present invention will become apparent from the description of the present specification and the accompanying drawings.

[0009]

Of the inventions disclosed by the present application, a representative one will be briefly described below.
It is as follows.

That is, in a job dispatch apparatus of a parallel computer system in which a plurality of CPUs load-balance and process a plurality of jobs, a CPU group in which a plurality of CPUs dedicated to the execution of a single job are connected in parallel, A group of input spool processing means for managing the execution waiting of a plurality of jobs executed by the CPU group; a resource competition management means for adjusting the competition of the resources used by the plurality of jobs executed by the CPU group; Output spool processing means for managing output waiting of a plurality of jobs to be executed, and among the plurality of jobs received by the input spool processing means, the resource competition management means judges that there is no resource conflict. Outputting the job by assigning the job to a CPU in the CPU group that has not executed the job and executing the job It is performed by the output spooling process means.

In the parallel computer system job dispatch apparatus, first, the input spool processing means receives a plurality of jobs to be executed by the parallel computer system.

Next, it is checked whether or not there is a vacant CPU that is not executing a job in the CPU group, and if the vacant CPU exists, a plurality of CPUs accepted by the input spool processing means are detected. The job in the waiting state before the execution is selected from the above jobs and the resources such as the magnetic disk device and the printer device used during the execution of the selected job are checked.

The resource conflict management means checks whether or not another job which uses the same resource as the resource used by the selected job during execution is currently being executed.

The resource conflict management means determines that another job using the same resource as the resource currently used by the selected job is not currently executed and resource conflict does not occur. If the selected job is a free C that is not executing a job of the CPU group,
When the job is assigned to the PU and executed, and the output processing is generated by the execution of the job, the output processing is performed by the output spool processing unit to adjust the competition of the output processing.

As described above, in the job dispatching device of the parallel computer system, the job is executed by the CPU dedicated to executing a single job.
The task switching process for switching to the task of another job does not occur, and when the number of CPUs forming the CPU group is increased, the number of jobs executed in parallel is proportional to the number of CPUs in the CPU group. Since the number of jobs increases, the linear scalability can be realized by the system expansion accompanying the increase of the job processing amount in the parallel computer system.

Further, in the job dispatching apparatus of the parallel computer system, the task switching processing does not occur in the CPU dedicated to the execution of a single job constituting the CPU group, so the CPU constituting the CPU group is designed. By simplifying, it is possible to install a large number of CPUs with good cost / performance in the parallel computer system.

As described above, according to the job dispatching device of the parallel computer system, CPUs dedicated to executing a single job are connected in parallel.
Since jobs are distributed and processed by groups, it is possible to realize efficient parallel execution of a plurality of jobs.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the present invention will be described in detail with reference to the drawings together with one embodiment.

In the job dispatching apparatus of the parallel computer system according to the present invention, a parallel processing of one embodiment in which a plurality of CPUs dedicated to the execution of a single job are connected in parallel to distribute and process the job A job dispatch device of a computer system will be described.

FIG. 1 is a diagram showing a schematic configuration of a job dispatch device of a parallel computer system of this embodiment. In FIG. 1, 1 is a job management CPU, 2 is an input spool, 3 is an output spool, 4 is a terminal device, and 5
Reference numeral 7 is a magnetic disk device, 8 is a printer device, 9 to 12 are jobs, 13 to 16 are job execution CPUs, and 17 is an output job.

As shown in FIG. 1, the job dispatching device of the parallel computer system according to this embodiment includes a job management CPU 1, an input spool 2, and an output spool 3.
, The terminal device 4, the magnetic disk devices 5 to 7, the printer device 8, the jobs 9 to 12, and the job execution CPU 13
16 to 16 and an output job 17.

Further, as shown in FIG. 1, in the job dispatching apparatus of the parallel computer system of this embodiment, the job management CPU 1 is connected to a terminal device 4, magnetic disk devices 5 to 7, and a printer device 8. And
Further, the job management CPU 1 has a plurality of jobs 9-12.
And an output spool 3 that receives the output processing of the output job 17.

In FIG. 1, DISK1, DISK2 and DISK3 are the names of the magnetic disk devices 5 to 7, respectively, and PRT1 is the name of the printer device 8, SORT1 and DRT.
BPRT1, SORT2, and DBACCS are the names of jobs 9 to 12, respectively.

As shown in FIG. 1, in the job dispatching apparatus of the parallel computer system according to the present embodiment, the job execution CPUs 13 to 16 are CPU groups in which a plurality of CPUs dedicated to the execution of a single job are connected in parallel. The job management CPU 1 is formed and the job execution CPU 1
Jobs 9 to 12 are assigned to 3 to 16.

Although not shown in FIG. 1, each of the plurality of job executing CPUs 13 to 16 forming the CPU group has a busy bit indicating whether or not the CPU is executing a job. ing.

FIG. 2 is a flow chart showing the processing procedure of the input spool processing of the job dispatching apparatus of the parallel computer system of this embodiment.

FIG. 3 is a flow chart showing the processing procedure of the device conflict management processing of the job dispatching apparatus of the parallel computer system of this embodiment.

FIG. 4 is a flowchart showing the processing procedure of the output spooling process of the job dispatching device of the parallel computer system of this embodiment.

Below, in the job dispatching apparatus of the parallel computer system of this embodiment, a plurality of jobs 9 to 12 are input from the terminal device 4 and the job execution C is executed.
Execute multiple jobs 9-12 with PUs 13-16,
A processing procedure for outputting the execution result to the terminal device 4 or the printer device 8 will be described.

The job management CPU 1 of the job dispatch device of the parallel computer system of this embodiment is
When a plurality of jobs 9 to 12 are input from the terminal device 4,
The input spool process, the device conflict management process, and the output spool process are executed in parallel in accordance with the control program written in the storage unit of the job management CPU 1. Here, the following procedure will be described in the order of job execution.

First, the job management CPU of the job dispatch device of the parallel computer system of this embodiment.
The job No. 1 inputs the jobs 9 to 12 from the terminal device 4 and carries out the input spool processing in the following procedure.

As shown in FIG. 2, in the input spool processing of the job dispatching apparatus of the parallel computer system according to the present embodiment, the job management CPU 1 executes a plurality of jobs 9 to 9 input from the terminal 4 in the processing of step 18. Twelve are accepted.

There are two types of jobs 9 to 12: a batch job by submitting from the terminal device 4 and a transaction for making an inquiry to the database, both of which are accepted here.

In the input spool processing of the job dispatching device of the parallel computer system of this embodiment, the priority of the jobs 9 to 12 accepted is the order of the job input time, and the number of jobs that can be accepted by the input spool processing is , The number of entries in the device conflict management table for registering the received job.

Next, in the processing of step 19, the plurality of jobs 9 to 12 accepted in the processing of step 18 are executed.
Is registered in the device conflict management table.

FIG. 5 is a diagram showing a schematic configuration of the device conflict management table after the input spool processing of the job dispatching apparatus of the parallel computer system of this embodiment. In FIG. 5, 28 is a device conflict management table after input spool processing, 29 is a JOB name registration field, and 30
Is a registration column of JOB-ID, 31 is a registration column of execution STATUS, 32 is a registration column of used device, 72 is a registration column of registration time, 33, 37, 42 and 46 are JOB names, 34,
38, 43 and 47 are JOB-ID, 35, 39 and 44.
And 48 execute STATUS, 36, 40, 45 and 4
9 is a device used, and 73, 74, 75 and 76 are registration times.

As shown in FIG. 5, the device conflict management table after the input spool processing of the job dispatching apparatus of the parallel computer system of this embodiment has a JOB name registration field 29, a JOB-ID registration field 30, and Execution S
Registration column 31 of TATUS and registration column 32 of used device
And a registration field 72 for registration time and JOB names 33, 37, 4
2 and 46 and JOB-ID 34, 38, 43 and 47
And execution statuses 35, 39, 44 and 48, used devices 36, 40, 45 and 49, and registration time 7
3, 74, 75 and 76.

Further, as shown in FIG. 5, in the device conflict management table after the input spool processing of the job dispatching apparatus of the parallel computer system of this embodiment,
JOB name 33 and JOB name 4 of the job to be sorted
2, SORT1 and SORT2, and DBPRT as the JOB name 37 of the job for printing the contents of the database
1. DBACCS is registered in the JOB name registration field 29 as the JOB name 46 of the job for searching the database.

Further, in the device conflict management table 28 after the input spool processing of the job dispatching device of the parallel computer system of this embodiment, the execution ST
Execution of the ATUS registration field 31 STATUS 35, 39,
"WAIT" of 44 and 48 is the registration field 29 of the JOB name.
SORT1, DBPRT1, SORT2 and DBA registered as JOB names 33, 37, 42 and 46 in
Each CCS job is in a waiting state before execution.

The used device registration column 32 of the device conflict management table 28 after the input spool processing shows the device name used by the job when the job registered in the job name registration column 29 is executed. .

After the input spool processing, the JOB is registered by the registration column 32 of the used device of the device conflict management table 28.
When the SORT1 which is the job registered as the name 33 is executed, it is understood that the SORT1 uses the DISK1 which is the magnetic disk device 5 indicated by the used device 36.

Further, as in the case of SORT1, J
DBPR that is a job registered as OB name 37
When T1 is executed, DBPRT1 uses DISK2 which is the magnetic disk device 6 shown in the used device 40, and when SORT2 which is a job registered as the JOB name 42 is executed, SORT2 uses the used device 4
DISK1 which is the magnetic disk device 5 shown in FIG.
And executes the job DBACCS registered as the JOB name 46, the DBACCS is the magnetic disk device 7 indicated by the used device 49, D.
It turns out that ISK3 is used.

As described above, the used device registration column 3 of the device conflict management table 28 after the input spool processing
Used devices 36, 40, 45 and 49 registered in No. 2
By checking, it is possible to check whether or not a device conflict occurs when a plurality of jobs of JOB names 33, 37, 42 and 46 registered in the JOB name registration field 29 are executed in parallel. .

For example, if SORT1 registered in JOB name 33 of JOB name registration field 29 of device conflict management table 28 after input spool processing and SORT2 registered in JOB name 42 are executed at the same time, the device used It can be seen that device competition occurs in DISK 1 shown in 36 and 45, that is, in the magnetic disk device 5.

In the job dispatching apparatus of the parallel computer system according to the present embodiment, the processing procedure of the device conflict management process for checking the device conflict during the job dispatch process will be described below.

The device conflict management processing of the job dispatching device of the parallel computer system of this embodiment is performed by starting up the parallel computer system of this embodiment (IPL: Initial Program L).
It shall be set in advance so that it will automatically start immediately after (oad).

In the device conflict management processing of the job dispatching apparatus of the parallel computer system of this embodiment, first, in the processing of step 20, a plurality of job executions C are executed.
The busy bits indicating the job execution status of the job executing CPUs 13 to 16 are checked to see if there are free CPUs that are not executing jobs in the PUs 13 to 16.

If the job execution C is performed in the process of step 20,
As a result of checking the busy bits of the PUs 13 to 16, if there is a job executing CPU for which the busy bit is off, it is determined that the job executing CPU is a free CPU that is not executing the job, and step 21 Go to processing.

In the processing of step 21, the device conflict management table 28 after the input spool processing shown in FIG. 5 is searched, and the content of the registration column 31 of the execution STATUS is "WAI".
Find the job that has "T".

Next, the registration column 32 of the used device of the job for which the content of the registration column 31 of the retrieved execution STATUS is "WAIT" is read, and the execution STATUS of another job using the read used device is read. Of the same device as the used device of the job whose execution status is “EX” indicating that the job is being executed and whose searched execution status is “WAIT”. The entire device conflict management table 28 is searched and checked as to whether or not the job being used exists.

In the device conflict management table 28 of the job dispatching device of the parallel computer system of the present embodiment, in the case of SORT1 which is the job registered in the JOB name 33, the execution STATUS 35 is "WAIT", Since there is no job in which the content of the registration column 31 of the execution status is “EX”, there is no job with the job name 33, SOR.
T1 is treated as a dispatch target job, and the process proceeds to step 22.

In the process of step 22, the content of the execution STATUS 35 of the registration column 31 of the execution STATUS of the device conflict management table 28 is changed to "EX", and the job of SORT1 is assigned to the empty CPU in the process of step 23. .

In the processing of step 21, if the execution STA
Execution S of a job whose TUS is "WAIT" and which uses the same device as the device used by the job
When there are a plurality of jobs whose TATUS is not “EX”, the contents of the registration field 72 of the registration time of the device conflict management table 28 are checked and the oldest job in time is selected as the dispatch target job.

As a result of checking the busy bits of the job execution CPUs 13 to 16 in the process of step 20, if there is no job execution CPU whose busy bit is off, or in the process of step 21, the execution status is At the same time as "WAIT", the execution status of the job that uses the same device as the device being used is "EX"
If no non-existent job exists, job 9 to
The device competition management process is terminated without allocating the 12 job execution CPUs 13 to 16.

In the job dispatching apparatus of the parallel computer system of the present embodiment, the necessary resources of the job are defined as a device for the sake of simplicity. However, in order to manage resource competition more finely, a magnetic disk is used. More efficient operation becomes possible if resource competition is managed in units of data sets stored in the devices 5 to 7 or in units of members of the data sets.

FIG. 6 shows the result of performing the device conflict management processing for the received jobs 9 to 12 in the job dispatching apparatus of the parallel computer system of the present embodiment as described above.

FIG. 6 is a diagram showing a schematic configuration of the device conflict management table after the device conflict management process of the job dispatching apparatus of the parallel computer system of this embodiment. In FIG. 6, 50 is a device conflict management table after the device conflict management process, 51 is a JOB name registration column, 52 is a JOB-ID registration column, and 53 is an execution STAT.
US registration column, 54 is a device registration column, 77 is a registration time registration column, and 55, 59, 64 and 68 are JOB.
First name, 56, 60, 65 and 69 are JOB-ID, 57,
61, 66 and 70 are execution statuses, 58, 62,
67 and 71 are used devices, 78, 79, 80 and 8
1 is the registration time.

As shown in FIG. 6, the device conflict management table after the device conflict management process of the job dispatching apparatus of the parallel computer system of this embodiment is JO.
B name registration column 51, JOB-ID registration column 52, execution STATUS registration column 53, device registration column 54, registration time registration column 77, JOB name 55, 5
9, 64 and 68, JOB-ID 56, 60, 65 and 69, and execution STATUS 57, 61, 66 and 70
And the use devices 58, 62, 67 and 71, and the registration times 78, 79, 80 and 81.

Further, as shown in FIG. 6, in the device conflict management table after the device conflict management process of the job dispatching apparatus of the parallel computer system of the present embodiment, the execution STATUS of the execution STATUS registration field 53 is registered.
“EX” in 57, 61, and 70 is a JOB name registration field 5
1 shows that each job of SORT1, DBPRT1 and DBACCS registered as JOB names 55, 59 and 68 is in a running state.

As shown in FIG. 6, in the device conflict management table 50 after the device conflict management processing of the job dispatching device of the parallel computer system of this embodiment, SORT2, which is the job of job name 64, is the job of job name 55. SORT1 use device 58
Since DISK1, which is the same device as the device indicated by, that is, the magnetic disk device 5 is used, it is considered that there is device conflict, and SORT2 execution STATUS66
Is held as "WAIT", and SORT2 is not a dispatch target job.

Since the dispatched job is executed by the assigned job execution CPU by itself, it is executed without performing the task switching process for switching to the task of another job.

When the jobs executed by the job execution CPUs 13 to 16 perform output processing for outputting results, the jobs are output to the output spool for adjusting the competition of the output processing when a plurality of output processings occur simultaneously. It will be in the output waiting state later.

As described above, in the job dispatching apparatus of the parallel computer system according to the present embodiment, all jobs accepted by the input spool process are connected in parallel in advance unless device competition with other jobs occurs. The prepared job execution CPUs 13 to 16 can be executed in parallel.

The processing procedure of the output spooling process for adjusting the competition of the output processes of the executed jobs in the job dispatching device of the parallel computer system of this embodiment will be described below.

As shown in FIG. 4, in the output spool processing of the job dispatching apparatus of the parallel computer system of this embodiment, first, in step 25, it is determined whether or not a job exists in the output spool.

As a result of checking whether or not the job exists in the output spool in the process of step 25, if the job exists in the output spool, the process proceeds to step 26 to execute the job output process. To do.

After the output processing of the job existing in the output spool is executed in the processing of step 26, step 2
7, the entry relating to the job that has executed the output process is deleted from the device conflict management table 50 after the device conflict management process.

As a result of checking whether or not the job exists in the output spool in the process of step 25, if the job does not exist in the output spool, the output spool process is ended.

As described above, in the job dispatching apparatus of the parallel computer system of this embodiment, the dispatched job is the assigned job execution CP.
Since it is executed without performing the task switching process in U, it is possible to eliminate the overhead of the task switching process, and since the task switching process does not occur, the job execution CPU design is simplified and the cost / performance is improved. It is possible to install a large number of good job execution CPUs.

In the job dispatching device of the parallel computer system of this embodiment, the number of job executing CPUs shown in FIG.
The number of PUs may be further increased, and C that executes the job
When the number of job executing CPUs forming the PU group is increased, the number of jobs that can be executed in parallel increases in proportion to the number of job executing CPUs. Therefore, it is possible to realize linear scalability of the parallel computer system. It is possible.

As described above, according to the job dispatching apparatus of the parallel computer system of the present embodiment, the CPU dedicated to executing a single job is load-balanced and processed by the group of CPUs connected in parallel. It is possible to realize efficient parallel execution of multiple jobs.

Although the present invention has been specifically described based on the above embodiment, the present invention is not limited to the above embodiment, and various modifications can be made without departing from the scope of the invention. Of course.

[0073]

The effects obtained by the representative ones of the inventions disclosed in the present application will be briefly described as follows.

That is, since the jobs dedicated to executing a single job are distributed and processed by the CPU group in which CPUs are connected in parallel, it is possible to realize efficient parallel execution of a plurality of jobs.

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of a job dispatch device of a parallel computer system according to an embodiment.

FIG. 2 is a flowchart showing a processing procedure of input spool processing of a job dispatching device of the parallel computer system of the present embodiment.

FIG. 3 is a flowchart showing a processing procedure of device conflict management processing of the job dispatching apparatus of the parallel computer system of the present embodiment.

FIG. 4 is a flowchart showing a processing procedure of output spool processing of the job dispatching apparatus of the parallel computer system of the present embodiment.

FIG. 5 is a diagram showing a schematic configuration of a device conflict management table after input spool processing of the job dispatching device of the parallel computer system of the present embodiment.

FIG. 6 is a diagram showing a schematic configuration of a device conflict management table after device conflict management processing of the job dispatching device of the parallel computer system of the present embodiment.

[Explanation of symbols]

1 ... Job management CPU, 2 ... Input spool, 3 ... Output spool, 4 ... Terminal device, 5-7 ... Magnetic disk device, 8
... printer device, 9-12 ... job, 13-16 ... job execution CPU, 17 ... output job, 28 ... device conflict management table after input spool processing, 29 ... JOB name registration column, 30 ... JOB-ID registration Column, 31 ... Execution ST
ATUS registration field, 32 ... Device usage field, 72
... Registration column for registration time, 33, 37, 42 and 46 ... JO
B name, 34, 38, 43 and 47 ... JOB-ID, 3
5, 39, 44 and 48 ... Execution STATUS, 36, 4
0, 45 and 49 ... Device used, 73, 74, 75 and 76 ... Registration time, 50 ... Device conflict management table after device conflict management processing, 51 ... JOB name registration column, 5
2 ... JOB-ID registration column, 53 ... Execution STATUS registration column, 54 ... Device registration column, 77 ... Registration time registration column, 55, 59, 64 and 68 ... JOB name, 5
6, 60, 65 and 69 ... JOB-ID, 57, 61,
66 and 70 ... Execution STATUS, 58, 62, 67 and 71 ... Device used, 78, 79, 80 and 81 ... Registration time.

Claims (1)

[Claims] 1. A job dispatching apparatus for a parallel computer system, wherein a plurality of CPUs load-balance and process a plurality of jobs, and a CPU group in which a plurality of CPUs dedicated to executing a single job are connected in parallel; Input spool processing means for managing execution waiting of a plurality of jobs executed in a group, resource competition management means for adjusting competition of resources used by a plurality of jobs executed in the CPU group, and the CPU
Output spool processing means for managing the output waiting of a plurality of jobs executed in a group, and of the plurality of jobs accepted by the input spool processing means, the resource competition management means determines that there is no resource conflict The job
A job dispatching apparatus for a parallel computer system, characterized in that a job in the CPU group is assigned to a CPU that is not executing the job, and the output processing of the job is performed by the output spool processing means.