JPH09160889A - Control method for multiprocessor system - Google Patents

Abstract

(57) Abstract: A load balancing is optimized in a function distributed multiprocessor system. SOLUTION: A functional processor group 104 that exclusively executes a synchronous channel job for transferring data between a cache memory 119, a host computer 101, and a channel control device 102, and data between a cache memory 119 and drives 120 to 122. Functional processor group 111 that exclusively executes synchronous / asynchronous command jobs for transfer
And a configuration management job for managing the configuration of the disk controller 103 in the disk controller 103 of the function-distributed multiprocessor system including the shared memory 118,
The execution of the cache management job that manages the free space on the cache memory 119 is not fixed to any of the specific functional processor groups 104 and 111, but is executed by the functional processor group with the lower load.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a control technology for a multiprocessor system, and is particularly effective when applied to a function distributed multiprocessor system in which a functional processor group for executing the job is fixed for each job. Regarding technology.

[0002]

2. Description of the Related Art For example, Iwanami Shoten, May 2, 1990.
Published on 5th, "Iwanami Information Science Dictionary" P438-P439,
In the field of information processing technology, a plurality of processors (processors) are operated in parallel as described in the literatures such as the above, so that high performance, excellent information in terms of reliability, expandability, etc. A multiprocessor system capable of constructing a processing system is known. In such a multiprocessor system, it is essential to improve the processing efficiency of the system to maintain the load evenly among the processors.

As load balancing control in a conventional function-distributed multiprocessor system, for example, a function processor group is fixed for each job as described in "Multiprocessor system" of Japanese Patent Laid-Open No. 6-35871. However, there is known a technique in which, when a job has a start request, when the job is executed by a processor in a fixed functional processor group, the load balance in the functional processor group is controlled.

[0004]

In the above-mentioned prior art,
The load balance control between the functional processor groups is not considered. That is, in the above-described conventional technology, since the functional processor group that executes even a job that does not belong to any functional processor group is fixed, the processor in the fixed functional processor group when there is a request to start the job. The job is executed by performing load balance control between them. Even if the load of the processor belonging to the functional processor group is higher than the load of the processor belonging to the other functional processor group, the job must be started, and thus the functional processor group becomes a bottleneck of the system. As described above, the above-mentioned conventional technology has a technical problem that the performance of the system is deteriorated because the load distribution and the load balance between the processor groups in the system are not taken into consideration.

An object of the present invention is, in a function distributed multiprocessor system, a multiprocessor system capable of improving the processing performance of the entire system by equalizing the load balance among a plurality of function processors or a group of function processors. To provide the control technology of.

[0006]

The present invention is applicable to any functional processor group in a multiprocessor system in which a functional processor or a functional processor group (execution processor group) for executing the job is fixed for each job. A non-job is a multiprocessor system in which a functional processor or a functional processor group is not fixed and is executed by an arbitrary functional processor or functional processor group. When there is a job execution request that does not belong to any functional processor or functional processor group, the load is set so that the functional processor or functional processor group with the lightest load among the functional processor or functional processor groups in the system executes the job. Control balance.

In order to realize the above load balance control, the following four means can be used as an example. That is, (1) means for starting a job that does not belong to any functional processor group in any functional processor group, (2) means for managing the load of each functional processor group, (3) load of each functional processor group. It is a means for making a determination and determining a functional processor group having a low load for activating the job, and (4) a means for activating the job with the functional processor group determined by the means of (3) above. By means of (1), the functional processor groups for activating jobs that do not belong to any of the functional processor groups are distributed, and by means of (2), (3) and (4), the functional processor groups in the system are The load balance is equalized and the system performance is improved.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a conceptual diagram showing an example of a configuration of a cached disk subsystem in which a method for controlling a multiprocessor system according to an embodiment of the present invention is implemented. In FIG. 1, the disk control device 103 is connected to the host computer 101 via the channel control device 102 on the upper side, and on the lower side, a drive 120, which is a storage device including a magnetic disk or the like as a recording medium. 121 and the drive 122 are connected. The disk controller 103 includes drives 120, 1
Data is read or written on the terminals 21 and 122 in response to a request from the host computer 101. The data transfer between the host computer 101 and the drives 120, 121, 122 is controlled by the channel control unit 105 and the channel control unit 108 built in the disk control device 103.
And a disk control unit 112 and a disk control unit 115.

These control units are internally provided with a processor 106, a processor 109, a processor 113, a processor 116, a local memory 107, a local memory 110, a local memory 114 and a local memory 1, respectively.
17 and a shared memory 118 for common control between control units
have. The channel control units 105 and 108 accept requests from the host computer 101 via the channel control device 102, and transfer data between the host computer 101 and the cache memory 119. The disk control units 112 and 115 transfer data between the cache memory 119 and the drives 120, 121 and 122. A request for data transfer from the channel control units 105 and 108 to the disk control units 112 and 115 and a data transfer completion report from the disk control units 112 and 115 to the channel control units 105 and 108 are performed via the shared memory 118.

FIG. 2 is a conceptual diagram showing an example of the structure of a table on the shared memory 118. The load information table 201 includes processors 106, 109, 113 and 116.
Load factor 202, load factor 203, load factor 2 for each of the
04 and load factor 205 are stored. The job management table 206 (Job Control Block: hereinafter abbreviated as JCB) has a one-to-one correspondence with each job, and has a JCB ID 207a that is identification information of the job,
Information such as a job activation request time 207b, a status 207c of the job, an executing processor number 207d indicating a processor executing the job, and the like are stored. The ready job queue 208 includes a processor 106,
The JCB 206 is queued in the job execution waiting queue 209, the execution waiting queue 210, the execution waiting queue 211, and the execution waiting queue 212 in a one-to-one correspondence with each of the 109, 113, and 116. The configuration information table 213 is a table for managing the configurations of the disk control device 103 and the drives 120, 121, 122.

FIG. 3 is a conceptual diagram showing an example of the configuration of various jobs executed in the disk control device 103. The job configuration in the disk control device 103 according to the present embodiment is the synchronous channel job 3 for transferring data between the host computer 101 and the cache memory 119.
01, a synchronous command job 302 for transferring data between the cache memory 119 and the drives 120, 121, 122 according to the instruction of the synchronous channel job 301, not the instruction of the synchronous channel job 301, but the cache memory 119
Asynchronous command job 30 that starts depending on the amount of data above
3, a configuration management job 304 that manages the configuration of the disk control device 103, and a cache management job 305 that manages free space on the cache memory 119.

As described above, each job is provided with a JCBID 207a of 4 digits (2 bytes) in hexadecimal. As an example, the JCB ID 207a of each job is “5800” for the synchronous channel job 301, “3C00” for the synchronous command job 302, “3800” for the asynchronous command job 303, and the configuration management job 30.
4 is assigned “2500”, and the cache management job 305 is assigned “2400”.

The processing contents of each of the above-described various jobs will be described.

The synchronous channel job 301 receives a request from the host computer 101 via the channel control device 102, and at the time of a read request, checks whether the requested data is in the cache memory 119,
When it is in the cache memory 119, the data in the cache memory 119 is transferred to the host computer 101, and when there is no such data, the disk control unit 112, 1
15 the stage of the data from the drives 120, 121, 122 to the cache memory 119 (stage: the drives 120, 121, 122 to the cache memory 1
19) operation, and when the data is transferred to the cache memory 119, the data in the cache memory 119 is transferred to the host computer 101. When a write request is made, it is checked whether old data corresponding to the write request data exists in the cache memory 119, and when it is in the cache memory 119, the write data is transferred from the host computer 101 to the cache memory 119, and then in the cache memory 119. If not, drive 1 to disk controller 112, 115
20, 121, 122 make a stage request of the data to the cache memory 119, and the cache memory 1
When the data is transferred to the cache memory 19, the write data is transferred from the host computer 101 to the cache memory 119.

The sync command job 302 staging the data requested by the sync channel job 301,
Alternatively, the requested data is destaged (destage: cache memory 119 to drives 120, 121, 1
22) Data transfer operation).

The asynchronous command job 303 performs destaging if there is undestaged data in the cache memory 119 even if there is no request from the synchronous channel job 301, or the host computer 1
The data stage that is likely to be requested from 01 is performed.

The configuration management job 304 manages the configurations of the disk controller 103 and the drives 120, 121, 122 with the configuration information table 213 in the shared memory 118, updates the table when a configuration change occurs,
The channel control units 105 and 108 and the disk control units 112 and 115 are notified of the configuration change occurrence.

The cache management job 305 is a job for managing the space in the cache memory 119. If there is data in the cache memory 119 that has already been written to the drive, it is stored in the cache memory 119. Frees the space used by the data. The configuration management job 304 and the cache management job 305 are activated at regular intervals.

The functional processor group of this embodiment will be described.

The functional processor group 104 capable of executing the functions of the synchronous channel job 301 is the channel control unit 10.
5, processor 109 inside processor 108
Consists of Functional processor group 11 capable of executing the functions of the synchronous command job 302 and the asynchronous command job 303
1 comprises processors 113 and 116 inside the disk control units 112 and 115. The configuration management job 304 and the cache management job 305 are jobs that do not belong to any of the functional processor groups 104 and 111.

FIG. 4 shows local memories 107, 110, 1
14 and 117 are conceptual diagrams showing an example of the contents of the load information creation table 401, and FIG. 5 is a flowchart showing an example of load information update processing. In the load information update processing, the start time is set to the job start time 4 before the job is started.
02 is set (step 501) and the job is started (step 502). When the job ends, the end time is set to the job end time 403 (step 503), and the job execution time which is the difference between the job end time 403 and the job start time 402 is set to the job execution time 404 (step 50).
4) and then the job execution time 4 is added to the job execution accumulated time 405.
04 is added (step 505). The job execution accumulated time 405 is updated until T seconds (for example, 1 second) has elapsed from the job execution accumulated start time 406 (step 506) (steps 501, 502, 503, 504, 50).
5) If the time has elapsed, the job execution integrated end time is set to the job execution integrated end time 407 (step 507), and the job execution integrated time 405 / (job execution integrated end time 40
7-job execution integrated start time 406) is set as the load ratio of the own processor in the load information table 201 (step 508), the job execution integrated time 405 is cleared (step 509), and the job execution integrated start time 406 is set.
Is reset (step 510) and the load information creation process is continued.

6 and 7 show a configuration management job 304 and a cache management job 305 under load balance control.
5 is a flowchart showing an example of a startup process of FIG. These jobs are jobs that are started at a fixed cycle, and the start request time is set to the own JC at the end of the job in step 703 of FIG. 7 described later.
The start request time 207b of B206 is set. Processors 106 and 109 of the functional processor groups 104 and 111,
113 and 116 are stored in J on the shared memory 118 at regular intervals.
The start request time 207b of the CB 206 is checked (step 601), and the JCB 20 that has reached the start request time
6 is determined to be the configuration management job 304 and the cache management job 305 by the JCB ID 207a (step 602), the processor with the lowest load factor is specified from the load information table 201 (step 60).
3) Queue the JCB in the ready job queue for the processor (step 604). In step 601, if there is no job that has reached the activation request time, nothing is done. Further, in step 602, the JCB which has reached the activation request time
Jobs in which the functional processors executed by 206 are fixed (sync channel job 301, sync command job 302,
In the case of the asynchronous command job 303), the ready job queue 20 of the processor 106, 109 or the processor 113, 116 belonging to the corresponding functional processor group
The job is queued in step 8 (step 60)
5).

Each processor checks whether or not there is a JCB queued in the ready job queue corresponding to its own processor (step 701), and if there is a queuing JCB, the job corresponding to that JCB is started (step 701). Step 702). When the job is finished, the next start time is set to the JCB start request time 207b corresponding to the job (step 703).

The operation of each functional processor group in the disk control device 103 in this embodiment has been described above. As described above, in this embodiment, the configuration management job 304 and the cache management job 305 that do not belong to either the functional processor group 104 or the functional processor group 111.
The functional processor group that activates is specified by the load balance control, and the functional processor group that is specified as having a smaller load is activated, so that the load balance between the functional processor group 104 and the functional processor group 111 can be made uniform.

That is, for example, in the prior art, when these jobs are fixed to the functional processor group 104 for convenience, if the requested data from the host computer is always in the cache memory 119, the load of the functional processor group 104 is increased. Is high and the load of the functional processor group 111 is low, the functional processor group 104 must be activated. When fixed to the functional processor group 111, when a large amount of data is destaged in the cache memory 119, the functional processor group 1
Although the load of 11 is high and the load of the functional processor group 104 is low, the functional processor group 111 must be activated. As a result, the load imbalance increases.

On the other hand, as in the present embodiment, a processor (functional processor group) having a low load is selected and the processor (functional processor group) is not fixed to any functional processor group. Management job 304
By activating the cache management job 305 and the cache management job 305, the load associated with the execution of the configuration management job 304 and the cache management job 305 is distributed, and each functional processor group (processor) during the data transfer process is not affected and the function distribution type It is possible to improve the system performance such as data transfer rate in the multiprocessor system.

Although the invention made by the present inventor has been specifically described based on the embodiments, the present invention is not limited to the above-described embodiments and can be variously modified without departing from the gist thereof. Needless to say, there is.

For example, the multiprocessor system is not limited to the disk control device exemplified in the above embodiment, but can be widely applied to general information processing technology of multiprocessor configuration.

[0030]

According to the control method of a multiprocessor system of the present invention, in a function distributed multiprocessor system, the load balance among a plurality of function processors or a group of function processors is made uniform to improve the processing performance of the entire system. The effect that it can be improved is obtained.

[Brief description of the drawings]

FIG. 1 is a conceptual diagram showing an example of a configuration of a disk subsystem with cache in which a method for controlling a multiprocessor system according to an embodiment of the present invention is implemented.

FIG. 2 is a conceptual diagram showing an example of control information used in a cached disk subsystem in which a method for controlling a multiprocessor system according to an embodiment of the present invention is implemented.

FIG. 3 is a conceptual diagram showing an example of a job configuration in a cached disk subsystem in which a method for controlling a multiprocessor system according to an embodiment of the present invention is implemented.

FIG. 4 is a conceptual diagram showing an example of control information used in a cached disk subsystem in which a method for controlling a multiprocessor system according to an embodiment of the present invention is implemented.

FIG. 5 is a flowchart showing an example of the operation of the cached disk subsystem in which the method for controlling a multiprocessor system according to an embodiment of the present invention is implemented.

FIG. 6 is a flowchart showing an example of the operation of the cached disk subsystem in which the method for controlling a multiprocessor system according to an embodiment of the present invention is implemented.

FIG. 7 is a flowchart showing an example of an operation of a disk subsystem with cache in which a control method for a multiprocessor system according to an embodiment of the present invention is implemented.

[Explanation of symbols]

101 ... Host computer, 102 ... Channel control device, 103 ... Disk control device, 104, 111 ... Functional processor group (processor group), 105, 108 ... Channel control unit, 112, 115 ... Disk control unit, 10
6, 109, 113, 116 ... Processor, 107, 1
10, 114, 117 ... Local memory, 118 ... Shared memory, 119 ... Cache memory, 120, 121,
122 ... Drive, 201 ... Load information table, 20
2, 203, 204, 205 ... Load factor, 206 ... Job management table (JCB), 207a ... JCBID, 20
7b ... Activation request time, 207c ... Status, 207d
... Processing processor number, 208 ... Ready job queue, 209, 210, 211, 212 ... Execution waiting queue, 213 ... Configuration information table, 301 ... Sync channel job (first job), 302 ... Sync command job (first Job), 303 ... Asynchronous command job (first job), 304 ... Configuration management job (second job), 305 ... Cache management job (second job).

Claims (1)

[Claims] 1. A plurality of processors or processor groups, and at least one first job whose execution is fixed to any of the plurality of processors or processor groups,
A method for controlling a multiprocessor system including at least one second job that can be executed by any of the processors or a group of processors, wherein the first and second jobs are executed when a request to start the job is issued. A method for controlling a multiprocessor system, which is identified, and when the activation request for the second job is made, the execution of the second job is assigned to the processor or processors having a smaller load.