JP3385407B2 - Non-stop program change method for multiprocessor system and multiprocessor system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a non-stop program changing method for a multiprocessor system and a multiprocessor system. More specifically, the present invention relates to a non-stop program changing method and a multiprocessor system for a multiprocessor system that can change a program executed by the processor group without stopping the entire processor group.

[0002]

2. Description of the Related Art Non-stop maintenance in which a main processor and a slave processor are provided and a backup system is switched so that the slave processor takes over the processing performed by the main processor, and then the main processor is stopped for maintenance. The method is described, for example, in JP-A-56-52460.

[0003]

According to the conventional non-stop maintenance method described above, it is possible to maintain the system portion belonging to the main processor without stopping the entire process. However, in order for the slave processor to take over the processing that was being performed by the master processor, the programs executed by the master processor and slave processors must be the same version. There is a problem that the above non-stop maintenance method cannot be applied. That is, the conventional technique has a problem that the processor group must be completely stopped when the program is upgraded. Therefore, an object of the present invention is to stop the entire processor group without stopping.
A non-stop program changing method of a multiprocessor system and a multiprocessor system capable of changing a program executed by a processor group.

[0004]

According to a first aspect of the present invention, a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to execute a job. In a multiprocessor system for processing, the plurality of processors are grouped into N clusters from cluster 1 to cluster N (N is a natural number less than or equal to 2 and less than or equal to the total number of processors), and a cluster to which the own processor belongs. A cluster-limited mode in which job distributed processing is performed only among processors belonging to the same cluster and a cluster-unlimited mode in which job distributed processing is performed among all processors are provided. When the job is distributed and the program is changed, first, the cluster-unlimited mode is used. From the job processing in the cluster-only mode to the job processing in the cluster-only mode, select one cluster when the migration is completed, and continue processing the jobs in the processor group of the non-selected cluster and accepting new jobs. Stop accepting new jobs in the processor group of the selected cluster, and when there are no jobs to be processed by the processor group of the selected cluster, replace the programs stored in the local memory of the processor groups of the selected cluster with the new programs. After changing the program of all processors in the selected cluster and restarting processing of new jobs and processing of jobs in the processors of the selected cluster, this is repeated for all clusters in order, and then the cluster-only mode Jobs in cluster-unlimited mode. Providing uninterrupted program changing the multiprocessor system back to the process, characterized in that return to normal operation.

In the non-stop program changing method for a multiprocessor system according to the first aspect, the processor group is divided into two or more clusters, and normally, jobs are distributed to all processors regardless of the clusters to change the program. In this case, first, move to a mode in which jobs are distributed only among processors in the same cluster, select one cluster after transition, and suspend processing in the selected cluster while continuing processing in non-selected clusters. To change the program of the processor of the selected cluster, restart the process after the change, repeat this for all clusters,
After the change of the program and the resumption of the processing are completed for all the clusters, the normal operation for distributed processing of the jobs is returned to all the processors regardless of the clusters. As a result, the program can be changed without completely stopping the operation of the multiprocessor system and without interfering with programs of different versions.

According to a second aspect, the present invention is a multiprocessor system in which a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to process a job. , Grouping the plurality of processors into N clusters from cluster 1 to cluster N (N is a natural number not less than 2 and not more than the total number of processors),
In addition, a cluster-limited mode in which the cluster group is divided into two sets, a first set and a second set, and job distribution processing is performed only between the cluster to which the own processor belongs and the processor group that belongs to the same cluster. A cluster non-limiting mode for performing distributed processing of jobs among a processor group belonging to the same cluster as the own processor and a group belonging to a cluster to which the own processor belongs ,
During normal operation, all clusters belong to one of the first group or the second group, and distributed processing of jobs is performed in the cluster-unlimited mode. When changing programs, first select one cluster and then deselect it. While continuing the distributed processing of jobs in the cluster-unlimited mode in the cluster, the job processing in the cluster-unlimited mode in the selected cluster shifts to the job processing in the cluster-limited mode, and when the migration is completed, the processor of the selected cluster The group stops accepting new jobs, and when there are no more jobs to be processed by the processors in the selected cluster, the programs stored in each local memory of the processors in the selected cluster are changed to new programs and selected. Select the processor of the cluster after completing the program change of all processors of the cluster Acceptance of a new job and job processing of the selected cluster is restarted, the affiliation of the selected cluster is changed from one of the first set and the second set to the other, and the job processing in the cluster-only mode of the selected cluster is performed. There is provided a non-stop program changing method for a multiprocessor system characterized by returning to processing of a job in a non-limited mode and repeating this in sequence for all clusters to return to normal operation.

In the non-stop program change method for a multiprocessor system according to the second aspect, the processor group is divided into two or more clusters, and normally, jobs are distributed to all processors regardless of the cluster to change the program. In this case, first, move to a mode in which jobs are distributed only between processors in the same cluster and processors in a predetermined cluster that can perform distributed processing, select one cluster when migration is possible, and continue processing in non-selected clusters. While suspending the processing in the selected cluster, changing the program of the processor in the selected cluster, restarting the processing after the change, and repeating this for all clusters, the change of the program and the restart of processing are completed for all clusters. Then, the job is distributed to all processors regardless of the cluster. To return to normal operation. As a result, the program can be changed without completely stopping the operation of the multiprocessor system, without interfering with programs of different versions, and performing distributed processing among the processors of different clusters.

In a third aspect, the present invention provides a multiprocessor system for implementing the non-stop program changing method according to the first aspect or the non-stop program changing method according to the second aspect.

In the multiprocessor system according to the third aspect, the above non-stop program changing method can be suitably implemented, so that it is not necessary to completely stop the operation when changing the program.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. The present invention is not limited to this.

First Embodiment FIG. 1 is a block diagram of a disk control device 200 according to the first embodiment of the present invention. This disk control device 200
Are processors 21 each having a local memory LM
1,212,213,214,221,222,223
And 224, and the control program storage drive 21
Service processor 21 having 6, 226 respectively
5, 225, and a shared memory 230 having a cache memory 231 and a control memory 232.

That is, this disk control device 200
Is eight processors 211, 212, 213, 21
A multiprocessor system having 4,221,222,223 and 224.

The eight processors 211, 212, 2
A data transfer control program and a maintenance support program are stored in the local memories LM 13, 214, 221, 222, 223 and 224. Each processor 211, 212, 213, 214, 221, 22
2, 223 and 224 execute the data transfer control program to perform data transfer control processing (that is, the disk control device 200 is a multiprocessor system). Further, the maintenance support program is executed, and non-stop maintenance processing is performed. Processors 211, 212, 221 and 222 of the processor group
Is a channel CP connected to the channel controller CC
Contains U. Therefore, these are called channel side processors. Further, the processors 213, 214, 223 and 224 include a drive CPU connected to a disk drive (hereinafter, simply referred to as a drive) D0. Therefore, these are called drive side processors. Further, the processor group is divided into two clusters. Processors 211, 212, 213 and 214 are “cluster 1” and processor 221,
222, 223 and 224 are “cluster 2”.
The service processors 215 and 225 are the maintenance terminals 2
It is connected to 40. In addition, the service processor 21
5 belongs to the "cluster 1", and the service processor 2
25 belongs to the "cluster 2".

The control memory 232 stores the cluster allocation information 500 as shown in FIG. The number of each processor is stored in the processor number area of this cluster allocation information 500 (2 in this embodiment).
11,212,213,214,221,222,22
3 and 224). In the belonging cluster number area, the cluster number of the cluster to which the corresponding processor belongs is stored (in this embodiment, the processor 21
1, 212, 213 and 214 are cluster numbers "1"
And the processors 221, 222, 223 and 22
4 is the cluster number "2"). Further, in the processor attribute area, information indicating the function of each processor in the distributed processing of the data transfer job is stored (in the present embodiment, the processors 211, 212, 221 and 22).
2 is a “channel CPU”, and processors 213, 2
14, 223 and 224 are "drive CPUs").

Further, the control memory 232 stores execution cluster instruction information 600 as shown in FIG. In the execution cluster indication area of this execution cluster indication information 600, "1 (cluster non-limitation mode)" is stored when either cluster 1 or cluster 2 is in a state where it can be used for data transfer. When only one of them can be used for data transfer, "2 (cluster limited mode)" is stored.

Further, in the control memory 232, I / O job execution control information 8 as shown in FIG. 4 is created corresponding to all the drives connected to the disk control device 200.
00 is stored. This I / O job execution control information 8
In the access command request area 00, the command type (read or write) of the access command and the drive number of the access target (I / O job execution control information 80
(Fixed for each 0), and the cylinder number, track number, and record number of the access target are stored. In the executable processor designation bitmap area, each processor 211, 212, 213, 214, 221, 222, 2
Information indicating whether or not 23 and 224 can execute the access command (“1” if executable, “0” if not executable) is stored. In the I / O request execution status area, information indicating the execution status of the access command for the drive (“0” if there is no execution request, drive CP
"1" is stored if U execution is waiting, "2" is stored if channel CPU is waiting, and "3" is stored if some processing is being executed.

Further, in the control memory 232, FIG.
The cluster operation mode instruction information 700 as shown in FIG. The cluster number (“cluster 1”, “cluster 2”) is stored in the cluster number area of the cluster operation mode instruction information 700. In the operation mode area, "1 (normal operation mode)" indicating a normal operation mode or "2 (sleep mode)" indicating an operation mode that does not accept an access command or the cluster 1
And cluster 2 store one of "3 (maintenance operation mode)" indicating that the data transfer control programs of different versions are being executed.

Next, the data transfer control process of the disk controller 200 will be described. When an access command is passed from the host computer (not shown) to the channel controller CC, the channel controller CC sends the access command to the cluster 1 (21) of the disk controller 200.
0) or one of the channel side processors 211, 212, 221, 222 of the cluster 2 (220). Here, it is assumed that the access command is a read request to the drive D0, and the access command is passed to the channel side processor 211 of the cluster 1 (210).

The channel side processor (211) to which the access command is passed is the I / O corresponding to the drive D0.
Only when "0 (no execution request)" is set in the I / O request execution status area of the job execution control information 800 (FIG. 4), "3 (currently executing)" is set in the I / O request execution status area. Is set, each information of the access command is stored in the access command request area, and "1 (drive CPU execution wait)" is set in the I / O request execution state area. Next, the channel side processor (211) refers to the execution cluster instruction information 600 (FIG. 3), and when the execution cluster instruction area is "1 (cluster non-limitation mode)", I / O job execution control Drive side processor number (213, 214, 223, 2 in the executable processor designation bitmap area of the information 800 (FIG. 4))
"1 (executable)" is set in the executability field corresponding to 24). On the other hand, when the execution cluster instruction area is “2 (cluster limited mode)”, the self-processor (here, the channel side in the executable processor designation bitmap area of the I / O job execution control information 800 (FIG. 4) is selected. "1 (executable)" is set only in the executability field corresponding to the drive side processor number (213, 214) that belongs to the same cluster (here, cluster 1) as the processor 211).

Drive side processor (213, 214,
223, 224) (here, the drive-side processor 223) refers to the I / O request execution status area of the I / O job execution control information 800 (FIG. 4), and If "(wait for drive CPU execution)" is set, refer to the executability field corresponding to the own processor number (223) in the executable processor designation bitmap area of the I / O job execution control information 800 (FIG. 4). However, only when "1 (executable)" is set there, "3 (running)" is set in the I / O request execution status area, and the information stored in the access command request area ( The data is read according to the drive number, cylinder number, track number, record number) and stored in the cache memory 232 of the shared memory 230, and the I / O job is executed. Control information 800 "2 (channel CPU execution wait)" in the I / O request execution state area (Fig. 4)
Set. Then, the execution cluster instruction information 600
(FIG. 3), when the execution cluster designation area is “1 (cluster non-limitation mode)”, among the executable processor designation bitmap areas of the I / O job execution control information 800 (FIG. 4). "1 (executable)" is set in the executability field corresponding to the channel side processor number (211, 212, 221, 222) of. On the other hand, when the execution cluster instruction area is “2 (cluster limited mode)”, the I / O job execution control information 8
00 (FIG. 4) in the executable processor designation bitmap area (the drive side processor 22
"1 (executable)" is set only in the executability field corresponding to the channel side processor numbers (221, 222) belonging to the same cluster (cluster 2) as 3).

Channel side processors (211, 212,
221 and 222) refers to the I / O request execution status area of the I / O job execution control information 800 (FIG. 4) and refers to “2 (Channel CPU execution wait) "is set only when" 1 (executable) "is set in the executability field corresponding to the own processor number (221) in the executable processor specification bitmap area. , I / O request execution status area is set to “3 (in execution)”, the corresponding data in the cache memory 232 is transferred to the channel controller CC, and I / O request execution status area is set to “0 (no execution request). ) ”Is set. The above is the read operation of the data transfer control process. The same is true at the time of writing, and the description is omitted.

Next, the non-stop maintenance process (control program change process) of the disk controller 200 will be described. Figure 6
FIG. 8 is a main flow diagram of non-stop maintenance processing (control program change processing). The maintenance person stores a new version of the data transfer control program in the control program storage drives 216 and 226 of the service processors 215 and 225 via the maintenance terminal 240 (step 11).
0). Next, the maintenance staff instructs the service processor 215 to start the control program change processing via the maintenance terminal 240 (step 120). Service processor 21
5 sets "2 (cluster limited mode)" to the execution cluster instruction information 600 (FIG. 3) (step 125).
Next, the service processor 215 executes the control program changing process of the cluster 1 (step 130). The control program changing process of the cluster 1 will be described later in detail with reference to FIG. While executing the control program changing process for the processors (211, 212, 213, 214) of the cluster 1, the processors (221, 222, 223, 224) of the cluster 2 execute the data transfer control program of the old version. continuing.

Next, the service processor 215 requests the service processor 225 to execute the control program changing process of the cluster 2 (step 140). The service processor 225 executes the control program changing process of the cluster 2 (step 145). The control program changing process of the cluster 2 will be described later in detail with reference to FIG. Cluster 2 processor (221,
222, 223, 224) while the control program changing process is being executed for the processors (21
1, 212, 213, 214) are executing a new version of the data transfer control program. When the service processor 215 confirms that the control program change processing of the cluster 2 is completed (step 150), the operation mode of each cluster in the cluster operation mode instruction information 700 (FIG. 5) is set to "1 (normal operation mode)". Set and set "1" in the execution cluster instruction information 600 (FIG. 3).
(Cluster non-limited mode) "is set (step 15
5). Then, a message indicating the completion of the control program change processing is displayed on the maintenance terminal 240 (step 16).
0).

FIG. 7 is a detailed flow chart of the control program changing process (130) of the cluster 1. Cluster 1
The processors (211, 212, 213, 214) of
By referring to the executable processor designation bitmap area of the I / O job execution control information 800 (FIG. 4) of all the drives (D0, D1, ...), “1” is simultaneously set for the processor of cluster 1 and the processor of cluster 2. It is confirmed that there is no job (cluster-unlimited job) that is "executable" (step 310). Since the execution cluster indication area of the execution cluster indication information 600 of FIG. 3 is set to “2 (cluster limited mode)” in step 125 of FIG. 6, the channel side processor (211, 21)
No. 2, 221, 222) receives a new access command from the channel control device CC, the cluster 1 or the cluster 1 or 2 is assigned to the executable processor designation bitmap area of the I / O job execution control information 800 (FIG. 4). Cluster 2
"1 (executable)" is set to only one of the processors. Therefore, the cluster-unlimited job will eventually disappear. Confirm this disappearance.

The service processor 215 then proceeds to FIG.
The operation mode of the cluster 1 of the cluster operation mode instruction information 700 is set to "2 (sleep mode)" (step 320). Cluster side channel processor (21
1, 212) notifies the channel controller CC that no new access command will be accepted when the operation mode of the cluster 1 in the cluster operation mode instruction information 700 is set to “2 (sleep mode)”. (Step 320). As a result, a new job to be executed by the processor of cluster 1 is not registered thereafter. Next, the processors of the cluster 1 (211, 212, 21
3, 214) refers to the executable processor designation bitmap area of the I / O job execution control information 800 (FIG. 4) of all drives (D0, D1, ...)
It is confirmed that there is no job (cluster limited job) that is "1 (executable)" for the processor (step 325). Then, it notifies the service processor 215 of the completion of suspension of its own processor (step 330).

Next, the service processor 215 executes the process shown in FIG.
The operation mode of the cluster 1 of the cluster operation mode instruction information 700 is set to "3 (maintenance operation mode)" (step 335). Next, the processor (21) that detects that the operation mode of the cluster 1 has been set to the maintenance operation mode
1, 212, 213, 214) request the service processor 215 to transfer the new version of the data transfer control program (step 340). In response to this, the service processor 215 transfers the new version of the data transfer control program to the processors (211, 212, 213, 214) of the cluster 1 and stores it in the respective local memories LM (step 345).
Cluster side channel side processors (211, 212)
Is a new version of the data transfer control program for all processors (211, 212, 21
(2, 214) confirming that it has been stored in the local memory LM, the channel controller CC is notified that the acceptance of a new access command is restarted (step 3).
50).

FIG. 8 is a detailed flow chart of the control program changing process (145) of the cluster 2. It should be noted that the jobs existing at the start of the control program changing process (145) of the cluster 2 are only the cluster limited jobs, and there are no cluster non-limited jobs. First, the service processor 225 uses the cluster operation mode instruction information 70 shown in FIG.
The operation mode of the cluster 2 of 0 is set to "2 (sleep mode)" (step 405). The channel side processors (221, 222) of the cluster 2 should not accept a new access command when the operation mode of the cluster 2 in the cluster operation mode instruction information 700 is set to “2 (sleep mode)”. The channel controller CC
(Step 415). As a result, a new job to be executed by the processor of cluster 2 will not be registered thereafter. Next, the processor (22
1, 222, 223, 224) are all drives (D
0, D1, ...) The I / O job execution control information 800 (FIG. 4) refers to the executable processor designation bitmap area, and the job (1 (executable)) for the processor of cluster 2 ( It is confirmed that there is no cluster limited job) (step 420). Then, the service processor 225 is notified of the completion of suspension of its own processor (step 425).

Next, the service processor 225 executes the process shown in FIG.
The operation mode of the cluster 2 in the cluster operation mode instruction information 700 is set to "3 (maintenance operation mode)" (step 426). Next, the processor (22) that detects that the operation mode of the cluster 2 has been set to the maintenance operation mode.
1, 222, 223, 224) requests the service processor 225 to transfer the new version of the data transfer control program (step 427). In response to this, the service processor 225 transfers the new version of the data transfer control program to the processors (221, 222, 223, 224) of the cluster 2 and stores it in the respective local memories LM (step 430).
Cluster 2 channel side processors (221, 222)
Indicates that the new version of the data transfer control program has all the processors (221, 222, 22) of the cluster 2.
3224), the channel controller CC is notified that the acceptance of a new access command is resumed (step 4).
40).

According to the above disk controller 200,
Processor group (211, 212, 213, 214, 22
1, 222, 223, and 224) can be completely upgraded, and the data transfer control program can be upgraded without interfering with the processors that execute the data transfer control programs of different versions.

-Second Embodiment- FIG. 9 is a block diagram of a disk controller 200 'according to a second embodiment of the present invention. This disk controller 20
0 ′ has the same configuration as the disk control device 200 of the first embodiment except that the processor group is divided into four clusters.

When the number of cluster divisions is three or more, it is possible to define a plurality of combinations of clusters capable of distributed processing of jobs among the processor groups. As for the combination of the clusters that can be distributed processed, the distributed processable cluster combination information 100 as shown in FIG. 10 is created and stored in the control memory 232. Distributed processable cluster combination information 1 in FIG.
In 00, if the processor group with the cluster number on the vertical axis and the processor group with the cluster number on the horizontal axis can perform distributed processing of jobs, "OK" is set in the field common to these cluster numbers. If job distributed processing is not possible, "NG" is set in the field common to these cluster numbers. Normally, when the processor group with the cluster number on the vertical axis and the processor group with the cluster number on the horizontal axis are executing the same version of the data transfer control program, "OK" is set and the data transfer control program of the same version is set. If it cannot be executed, "NG" is set. For example, when the processor groups of all the clusters are executing the same version of the data transfer control program, "OK" is set in all the fields as shown in FIG. On the other hand, the processor group of cluster 1 (211, 21
3) only has started executing the new version of the data transfer control program, and the processors (212, 214, 221, 223, 222, 222, 214, 221, 223, 222) of cluster 2 to cluster 4 have started.
224), when the old version of the data transfer control program is being executed, “NG” is set in the field corresponding to cluster 1 as shown in FIG. 11, and the field common to cluster 2 to cluster 4 is set. Is "O
K ”is set. In this way, the version of the data transfer control program executed by the processor group of each cluster changes at the time of maintenance operation, and accordingly, the setting value of the distributed processable cluster combination information 100 is dynamically changed accordingly. To do.

Then, in each processor, the execution cluster designation area of the execution cluster designation information 600 of FIG.
(Cluster limited mode) ”, the cluster to which the own processor belongs in the distributed processable cluster combination information 100 in the executable / non-executable field of the executable processor designation bitmap area of the I / O job execution control information 800 of FIG. Is set to "1 (executable)" only in the executability field corresponding to the processor belonging to the other cluster for which "OK" is set.

According to the disk control device 200 'described above, the version of the data transfer control program can be upgraded in finer cluster units. Further, jobs can be distributed and processed among processors having the same version of the data transfer control program.

FIG. 12 shows a diagram corresponding to FIG. 11 when the processor group is divided into five clusters. This distributed processable cluster combination information 100 'is the cluster 1
And the program change process of the processor of cluster 2 is completed, the program change process of cluster 3 is in the process of being executed, and the program change process of the processors of cluster 4 and cluster 5 is not yet started. There is. That is, since the processors of cluster 1 and cluster 2 are executing a new program, cluster 1 and cluster 2 are one set capable of distributed processing. Also, cluster 4 and cluster 5
, The cluster 4 and cluster 5 are the other set capable of distributed processing. Further, since the cluster 3 is under program change, it cannot be distributed to other clusters.

-Third Embodiment- Fig. 13 is a block diagram of a disk control device 200 "according to a third embodiment of the present invention. This disk control device 20
0 "is the processor (211, 212, 213, 21
4), the service processor (215) and the shared memory (2)
30) is connected with a bus 201. In this case, the processors (211, 21) connected to the bus 201
(2,213,214) is divided into appropriate clusters,
The data transfer control program can be upgraded by the same maintenance method as in the first and second embodiments.

[0036]

According to the non-stop program changing method of a multiprocessor system of the present invention, a program can be changed without completely stopping the operation of the multiprocessor system and without interfering with programs of different versions. Can be done.

According to the multiprocessor system of the present invention, there is no need to stop the operation for the purpose of changing the program.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a disk control device according to a first embodiment of the present invention.

FIG. 2 is a configuration diagram of cluster allocation information.

FIG. 3 is a configuration diagram of execution cluster instruction information.

FIG. 4 is a configuration diagram of I / O job execution control information.

FIG. 5 is a configuration diagram of cluster operation mode instruction information.

FIG. 6 is a main flow diagram of non-stop maintenance processing (control program change processing).

FIG. 7 is a detailed flowchart of a control program change process of cluster 1.

FIG. 8 is a detailed flowchart of a control program changing process of the cluster 2.

FIG. 9 is a configuration diagram of a disk control device according to a second embodiment of the present invention.

FIG. 10 is a configuration diagram of distributed processable cluster combination information.

FIG. 11 is another configuration diagram of distributed processable cluster combination information.

FIG. 12 is yet another configuration diagram of distributed processable cluster combination information.

FIG. 13 is a configuration diagram of a disk control device according to a third embodiment of the present invention.

[Explanation of symbols]

100, 100 'Distributed processable cluster combination information 200, 200', 200 "Disk controller 201 Bus 211, 212, 221, 222 Channel side processor 213, 214, 223, 224 Drive side processor 215, 225 Service processor 230 Shared memory 232 Control memory 240 Maintenance terminal 500 Cluster allocation information 600 Execution cluster instruction information 800 I / O job execution control information 700 Cluster operation mode instruction information CC channel controller D0, D1 Disk drive

─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Kenzo Kurihara 1099 Ozenji, Aso-ku, Kawasaki-shi, Kanagawa Inside Hitachi Systems Development Laboratory (56) Reference JP 5-324303 (JP, A) JP JP 7-210378 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 15/177 670 G06F 12/16 310 G06F 13/00 530 G06F 15/16 640

Claims (6)

(57) [Claims] 1. A multiprocessor system in which a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to process a job. To cluster N (N
Is a cluster limited mode in which N is divided into N clusters of 2 or more and a natural number less than or equal to the total number of processors), and job distribution processing is performed only between processors belonging to the same cluster as the own processor. And cluster non-restricted mode for distributed processing of jobs between all processors. During normal operation, distributed jobs are processed in cluster non-restricted mode, and during program change operation, first in cluster non-restricted mode. Job processing in the cluster-only mode, select one cluster when the migration is complete, and select while continuing to process jobs and accept new jobs in the processor groups of non-selected clusters. Stop accepting new jobs on cluster processors
When there are no more jobs to be processed by the processors in the selected cluster, change the programs stored in each local memory of the processors in the selected cluster to new programs, and when you have finished changing the programs in all processors in the selected cluster. After accepting a new job and processing a job in the processor group of the selected cluster and repeating this for all clusters in order,
A non-stop program changing method for a multiprocessor system, wherein the job processing in the cluster-only mode is returned to the job processing in the non-cluster-only mode to return to normal operation. 2. A multiprocessor system in which a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to process a job. To cluster N (N
Is a natural number less than or equal to 2 and less than or equal to the total number of processors), and is divided into N clusters, and the first cluster group is
A cluster-limited mode in which the job is distributed only between the cluster to which the self-processor belongs and the cluster to which the self-processor belongs, and the cluster to which the self-processor belongs. A cluster-unlimited mode is provided for distributed processing of jobs between a group to which a processor group belonging to the same cluster and a cluster to which the own processor belongs and a cluster non-limiting mode for performing job distributed processing between processor groups belonging to the same group. Clusters belong to either the first group or the second group, the jobs are distributed in the cluster non-restricted mode, and at the time of the program change operation, one cluster is first selected and the cluster non-restricted mode in the non-selected cluster is selected. Jobs in the cluster-unlimited mode in the selected cluster while continuing the distributed processing of jobs in The processing shifts to the job processing in the cluster-only mode, when the migration is completed, the new cluster's processor group stops accepting new jobs, and when the selected cluster's processor group has no jobs to process, the selected cluster Change the program stored in each local memory of each processor group to a new program respectively, and when the change of the programs of all the processors of the selected cluster is completed, accept the new job and the job of the processor group of the selected cluster. Restarts the processing, changes the affiliation of the selected cluster from one of the first group or the second group to the other, and returns the job processing in the cluster-only mode in the selected cluster to the job processing in the non-cluster-only mode. , Repeat this for all clusters in order to return to normal operation. And a non-stop program changing method for a multiprocessor system characterized by: 3. A multiprocessor system in which a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to process a job. To cluster N (N
Is a cluster allocating means for grouping into N clusters of 2 or more and a natural number less than the total number of processors) and a cluster for performing job distributed processing only between the cluster to which the own processor belongs and the processor group belonging to the same cluster. Distributed processing mode control means for controlling to perform distributed processing in either a limited mode or a cluster non-limited mode in which jobs are distributed among all processors; and in normal operation, the cluster non-limited mode To perform distributed processing of jobs, and at the time of program change operation, first shift from job processing in non-cluster mode to job processing in cluster-limited mode, select one cluster when the migration is completed, Continue processing jobs and accepting new jobs in the processors in the non-selected cluster. Stop accepting new jobs in the processor group of the selected cluster, and when there are no jobs to be processed by the processor group of the selected cluster, replace the programs stored in the local memories of the processor groups of the selected cluster with the new programs. After changing the program of all the processors of the selected cluster, accept the new job and process the jobs in the processor group of the selected cluster, and repeat this for all the clusters in order. And a program change processing unit that shifts the job processing in step 1 to the job processing in the cluster-unlimited mode and returns to normal operation. 4. A multiprocessor system in which a program is stored in each local memory of a plurality of processors and each processor executes a program stored in its own local memory to process a job, wherein the plurality of processors are cluster 1 To cluster N (N
Is a cluster assigning means for grouping N clusters up to 2 and a natural number less than the total number of processors) and a cluster capable of distributed processing for grouping the cluster groups into two sets, a first set and a second set. A cluster limitation mode in which job distribution processing is performed only between the combination means and the cluster to which the own processor belongs and the cluster to which the own processor belongs and the cluster to which the own processor belongs and the cluster to which the own processor belongs Distributed processing mode control means for controlling distributed processing in either a cluster-unlimited mode in which distributed processing of jobs is performed between processors belonging to the same group as the group to which the processing belongs, and during normal operation Assigns all clusters to either the first or second set, and When the program change operation is performed, first select one cluster and continue the job distributed processing in the cluster non-restricted mode in the non-selected cluster while continuing the job distributed processing in the non-selected cluster in the selected cluster. When the job processing is shifted to the job processing in the cluster-only mode, when the migration is completed, the acceptance of a new job by the processor group of the selected cluster is stopped, and when there are no jobs to be processed by the processor group of the selected cluster. The program stored in each local memory of the processor group of the selected cluster is changed to a new program, and when the change of the programs of all the processors of the selected cluster is completed, acceptance of a new job in the processor group of the selected cluster and Restart the job processing and select Change the job affiliation from one of the first set or the second set to the other, and then shift the job processing in the cluster-limited mode in the selected cluster to the job processing in the non-cluster-limited mode. A multiprocessor system comprising: a program change processing means for repeatedly repeating a cluster and returning to a normal operation. 5. A multiprocessor system in which a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to process a job, wherein two or more of the plurality of processors are provided. Cluster allocation information storing means for storing cluster allocation information indicating which cluster each processor belongs to in order to be grouped into each cluster, and a normal operation mode in which the processor processes jobs according to a program stored in the local memory. Whether to set the operation mode to either the sleep mode in which the processor suspends job processing to close the cluster or the maintenance operation mode in which the program stored in the local memory is changed to a new program Indicated by The operation mode instruction information storage means for storing the operation mode instruction information and the cluster limited mode in which the distributed processing of the job is performed only by the processor belonging to the same cluster as the self processor, or the job is executed between arbitrary processors regardless of the cluster. Cluster non-limitation mode in which the distributed processing is performed, or execution cluster instruction information storage means for storing execution cluster instruction information indicating in which distributed processing mode the job is processed, and program change start for instructing start of program change When the instruction means and the instruction to start the program change are instructed, the distributed processing mode is set to the cluster-only mode, an arbitrary cluster is selected, the operation mode of the cluster is set to the dormant mode, and the processor belonging to the cluster is set. When all of the Mode is set to maintenance operation mode, the program stored in the local memory of all processors belonging to the cluster is changed to a new program, this is repeated for all clusters, and the operation mode of all clusters at the completion To a normal operation mode and a distributed processing mode to a cluster non-limitation mode, and a program change processing means. 6. A multiprocessor system in which a program is stored in each local memory of a plurality of processors, and each processor executes a program stored in its own local memory to process a job, wherein two or more of the plurality of processors are provided. Cluster allocation information storing means for storing cluster allocation information indicating which cluster each processor belongs to in order to be grouped into each cluster, and a normal operation mode in which the processor processes jobs according to a program stored in the local memory. Whether to set the operation mode to either the sleep mode in which the processor suspends job processing to close the cluster or the maintenance operation mode in which the program stored in the local memory is changed to a new program Indicated by Operation mode instruction information storage means for storing operation mode instruction information, and a cluster limited mode in which distributed processing of jobs is performed only by processors belonging to the same cluster as the own processor, or belonging to the same cluster as the own processor Execution cluster instruction information that indicates in which distributed processing mode the job is to be processed by the processors of other clusters that allow distributed processing of jobs between the processor and its own processor Execution cluster instruction information storage means, program change start instruction means for instructing program change start, and when a program change start is instructed, an arbitrary cluster is selected and the distributed processing mode of the cluster is selected. Set the cluster-only mode, and The operation mode is set to the dormant mode, and when all the processors belonging to the cluster are in the dormant state, the operation mode of the cluster is set to the maintenance operation mode and stored in the local memory of all the processors belonging to the cluster. A program that changes the existing program to a new program, sets the distributed processing mode of the cluster to the cluster-unlimited mode, repeats this for all clusters, and sets the operation mode of all clusters to the normal operation mode at the time of completion. A multiprocessor system comprising: a change processing unit.