JP2853654B2 - Alternative processing method in tightly coupled 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 tightly coupled multiprocessor system comprising a plurality of processors.
More specifically, in a system in which a processor group is divided into several processor groups and a job is executed by specifying each processor group, a job in which a processor group in which a job cannot be executed due to a failure of all processors in the group is specified. The present invention relates to a technique for distributing the load of a replacement destination processor group when another processor group is substituted.

[0002]

2. Description of the Related Art Conventionally, load distribution in the case of a processor failure in a tightly coupled multiprocessor system has been performed for each processor constituting the system. For example, Japanese Patent Application Laid-Open No. 3-219360 discloses a multi-processor system having a plurality of processors.
An alternative processing capacity table is prepared for each processor to determine the load of each processor, and based on this, an alternative information table is created in the event of a failure of a certain processor. By performing the unit substitution, a decrease in the throughput of the system when a processor failure occurs is suppressed. In addition, no special consideration is given to reviewing the alternative processor once the alternative processor has been determined.

[0003]

By the way, in recent years, in order to improve the processing capability of a tightly coupled multiprocessor system, the number of connected processors has been increased from the conventional one. However, in such a system, in order to prevent performance degradation due to contention between main memory and buses, the processors in the system are grouped into a small number of processors and compared for each processor group. A large-capacity cache memory and a dispatcher for allocating tasks, which are the execution units of the program, are placed in groups so that multiple operations can be performed, and each processor group is made as close as possible to a conventional tightly coupled multiprocessor system ( For example, NEC
Memory-shared parallel system in a parallel ACOSPX7800 manufactured by FUJITSU LIMITED. And in such a system,
When a job is input to the system, a user specifies a processor group to be executed by the job, and the job is executed in the specified processor group.

In a tightly coupled multiprocessor system that executes a job by specifying a processor group as described above, when a processor failure occurs, another processor capable of executing a job is included in the processor group to which the failed processor belongs. In some cases, the processing is substituted by that processor, but when all the processors in the same processor group cannot execute the job due to a failure, it is necessary to substitute the processing of that processor group with another processor group, At this time, load imbalance between processor groups becomes a problem.

As described above, load distribution at the time of a processor failure in a conventional tightly coupled multiprocessor system is as follows.
This technique is performed in units of individual processors constituting the system. In the tightly coupled multiprocessor system described above, when a processor failure occurs, another job is added to a processor group to which the failed processor belongs. Can be applied to the load distribution in the processor group when there is a processor capable of executing the process. However, since the configuration manages load information and alternative information for each processor, when all processors in the same processor group cannot execute a job due to a failure, the processing of that processor group is performed by another processor group. It cannot be applied to load balancing in the case of substitution, and a special mechanism for that is required.

[0006] Therefore, an object of the present invention is to make the load of the remaining processor groups as equal as possible when the processing cannot be executed by one or more processor groups and the processing is replaced by the remaining processor groups. An object of the present invention is to provide an alternative processing method in a tightly coupled multiprocessor system that can prevent the throughput of the entire system from being reduced.

Another object of the present invention is to provide a tightly coupled multiprocessor system which executes a job by designating a processor group to which the present invention is applied. Is the number of processors belonging to each processor group and the number of processors that can execute jobs (that is, the number of available processors) in consideration of the fact that it is given that all processors in the system are available. ) Is to determine an alternative processor group based only on simple information such as).

Still another object of the present invention is to review an alternative processor group when an arbitrary processor state changes in an alternative processor group or the like (change from available to unavailable and vice versa). Accordingly, an object of the present invention is to provide an alternative processing method in a tightly-coupled multiprocessor system that enables an alternative processing according to the situation at each time.

[0009]

An alternative processing method in a tightly coupled multiprocessor system according to the present invention is to divide a plurality of processors sharing a main memory into several processor groups and execute a job by designating each processor group. Is characterized by including the following components. Processor number storage area; provided for each processor group in a one-to-one correspondence, and stores the number of processors belonging to the processor group. A usable processor number storage area; provided for each processor group in a one-to-one correspondence, and stores the number of processors capable of executing a job among the processors belonging to the processor group. Executable processor group number storage area; provided for each processor group in a one-to-one correspondence, and stores the number of another processor group that replaces the job specified by that processor group. Processor group management table initialization means: At the time of system initialization, the number of processors included in each processor group and the number of available processors among them are stored in a processor number storage area and an available processor number corresponding to each processor group. Store in the area,
Control is passed to the following processor group optimization means. ○ Table updating means: If a processor that cannot execute a job occurs during system operation, the value of the available processor number storage area corresponding to the processor group to which the processor belongs is decremented by 1 and Control is passed to the processor group optimizing means, and when a processor capable of executing a job occurs, the value of the available processor number storage area corresponding to the processor group to which the processor belongs is increased by 1 and Is passed to the processor group optimizing means. Processor group optimizing means; every time control is passed from the processor group management table initializing means or the table updating means, the processor group designation is performed for all processor groups in which the value of the available processor number storage area is 0. Is determined in the execution processor group number storage area corresponding to the substitution source processor group, and when a job exists in the system, the following job execution group changing means is controlled. hand over. ○ Job execution group changing means: for all processor groups, if the value of the available processor number storage area is not 0, the job assigned to another processor group among the jobs specified by the processor group is specified. In the case of a processor group in which the value of the available processor number storage area is 0, the job designated by the processor group is assigned to another processor group corresponding to the number stored in the execution processor group number storage area corresponding to the processor group. Assign to processor groups. ○ Job execution group selection means; When a job specified by a processor group is submitted, if the value of the available processor number storage area corresponding to the specified processor group is not 0, the job is assigned to the specified processor group and the job is specified. If the value of the available processor number storage area corresponding to the processor group is 0, the job is assigned to the processor group of the number stored in the execution processor group number storage area corresponding to the specified processor group.

Here, the processor group optimizing means assigns a job specifying a processor group having a value of the available processor number storage area of 0 to a processor group having a value of the available processor number storage area other than 0. For all combinations to be replaced in units, the load per available processor of the processor group after the replacement, in which the value of the available processor number storage area is not 0, is expressed by the number of input jobs specified by each processor group. The calculation is performed assuming that the number of processors is almost proportional to the number of processors stored in the storage area, and a processor group associated with a combination having the smallest load variation among all combinations is determined as an alternative processor group.

In the alternative processing method in the tightly coupled multiprocessor system having such a configuration, the processor group management table initializing means at system initialization initializes the number of processors included in each processor group and the number of processors usable therein. The processor number is stored in the processor number storage area and the available processor number storage area corresponding to each processor group, and control is passed to the processor group optimizing means, and the processor group optimizing means stores the available processor number storage area. For all processor groups having a value of 0, the processor group number for substituting the job specified by the processor group is determined and stored in the execution processor group number storage area corresponding to the processor group.

At this time, the processor group optimizing means assigns the job designating all the processor groups whose available processor number storage area value is 0 to the processor group whose available processor number storage area value is not 0. For all combinations to be replaced in units, the load per available processor of the processor group after the replacement, in which the value of the available processor number storage area is not 0, is expressed by the number of input jobs specified by each processor group. The calculation is performed on the assumption that the number of processors is almost proportional to the number of processors stored in the storage area, and an alternative processor group for a combination having the smallest load variation among all the combinations is determined.

When a job is specified by specifying a processor group after the initialization of the system, the job execution group selecting means selects the specified processor group if the value of the available processor number storage area corresponding to the specified processor group is not 0. A job is assigned to a processor group. If the value of the available processor number storage area corresponding to the specified processor group is 0, the job is assigned to the processor group of the number stored in the execution processor group number storage area corresponding to the specified processor group. Assign a job. As a result, a job that specifies a processor group in which all processors in the group are already unusable at the time of system startup is executed in the substitute destination processor group.

Next, when a processor that cannot execute the job occurs while the system is operating, or when a processor that is in a fault but recovers and can execute the job occurs. , The table updating means sets the value of the available processor number storage area corresponding to the processor group to which the processor belongs to 1
The control is passed to the processor group optimizing means by decreasing or increasing by 1, and the processor group optimizing means deletes the job specified by the processor group for all the processor groups in which the value of the available processor number storage area is 0. The number of the processor group to be replaced is determined and stored in the execution processor group number storage area corresponding to the processor group, and when there is a job in the system, the control is passed to the job execution group changing unit. Then, the job execution group changing means,
In the case of a processor group in which the value of the available processor number storage area is not 0 for all processor groups, a job assigned to another processor group among the jobs specified by the processor group is assigned to the processor group, and the available processor In the case of a processor group whose value of the number storage area is 0, the job specified by the processor group is assigned to another processor group corresponding to the number stored in the execution processor group number storage area corresponding to the processor group.

As a result, the following operation is performed by changing the processor state during the operation of the system.

(1) When all the processors of a certain processor group cannot execute a job due to a processor failure, a replacement destination processor group of this processor group and a replacement destination processor group of another processor group that is already unavailable. Is determined, and the job specifying the disabled processor group is changed to be executed in the alternative destination processor group.

(2) When at least one processor of the disabled processor group becomes available, the job specified by the processor group and assigned to the alternative processor group becomes the original processor group. Is changed to be executed by

(3) The number of available processors does not become zero when the number of available processors decreases by one, and the number of available processors does not become one when the number of available processors decreases by one. Even when the number increases or decreases, if there is a processor group in which all processors are unusable, the replacement destination processor group is reviewed.

The present invention is characterized in that the following table updating means is provided in place of the above table updating means. Table updating means: When a processor that cannot execute a job occurs during the operation of the system, the value of the available processor number storage area corresponding to the processor group to which the processor belongs is decremented by one. When the value becomes 0, the control is transferred to the processor group optimizing means. If a processor capable of executing a job occurs, the value of the available processor number storage area corresponding to the processor group to which the processor belongs is generated. Is incremented by 1 and when the value becomes 1, control is passed to the processor group optimizing means. The operation of the configuration having such a table updating means is the same as the previous configuration except that the review at the time (3) is not performed.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a schematic block diagram showing an example of a tightly coupled multiprocessor system to which the present invention is applied. The tightly-coupled multiprocessor system of this example includes a plurality of processors P11 to P1n,.
Pm1 to Pmn are replaced by m processor groups PG1 to P
Gm, and relatively large-capacity cache memories CM1 to CMm are provided for each of the processor groups PG1 to PGm. B is a system bus, and IOP is an input / output processor.

In this tightly coupled multiprocessor system, the main control of the whole system is O
A dispatcher for allocating a task, which is a unit of execution of a program, is provided for each processor group to enable multiplex operation.
PGm is made as close as possible to a conventional tightly coupled multiprocessor system. Further, when a job is input to the system, the user specifies a processor group to execute the job, and the OS basically assigns the job to the specified processor group. The assignment of a job to each of the processor groups PG11 to PGm is performed by registering job execution information in queues Q1 to Qm corresponding to each of the processor groups PG1 to PGm provided in the main storage device MM.
1 to PGm dispatchers Q1 to PGm
The job execution information is sequentially input from Qm, and the job is executed by the processors in the own group. If all processors in the own group cannot execute the job due to a failure during job execution, the job execution information is dequeued to the queues Q1 to Qm.

The present invention is not limited to the application to a tightly-coupled multiprocessor system of the type illustrated in FIG. 1, but any type of tightly-coupled multiprocessor that executes a job by specifying a processor group. Applicable to the system.

FIG. 2 is a block diagram showing one embodiment of an alternative processing method in the tightly coupled multiprocessor system of the present invention. The alternative processing method of this example includes a usable processor number storage area group 1, a processor number storage area group 2, an execution processor group number storage area group 3, a processor group management table initializing means 11, a table updating means 21 , A processor group optimizing unit 12, a job execution group changing unit 22, and a job execution group selecting unit 31. These are incorporated as part of the OS.

The processor number storage area group 2 includes processor number storage areas 2-1 to 2-m corresponding to the processor groups PG1 to PGm on a one-to-one basis. In the processor number storage areas 2-1 to 2-m, the number of processors belonging to the processor groups PG1 to PGm is stored by the processor group management table initialization means 11 when the system is initialized, as described later. .

The available processor number storage area group 1 is:
The processor groups PG1 to PGm are constituted by available processor number storage areas 1-1 to 1-m corresponding to the processor groups PG1 to PGm on a one-to-one basis. Storage area for number of usable processors 1-1
As described later, the processor group management table initializing means 11 stores, among the processors belonging to the processor groups PG1 to PGm, processors (executable processors) which can execute a job when the system is initialized. The number of processors (processors) is stored, and the value is changed by the table updating means 21 in accordance with a change in the processor state during system operation.

Execution processor group number storage area group 3
Are the execution processor group number storage areas 3-1 to 3 corresponding to the processor groups PG1 to PGm on a one-to-one basis.
-M. In each of the execution processor group number storage areas 3-1 to 3-m, as described later, the numbers of other processor groups which substitute the job specified by the processor group by the processor group optimizing means 12 are stored.

The processor group management table initialization means 11 is a means for storing initial information in the processor number storage area group 2 and the available processor number storage area group 1 at the time of system initialization. The processor group management table initialization means 11 refers to, for example, a configuration information table (not shown) describing the number of processors belonging to each of the processor groups PG1 to PGm and refers to a processor included in each of the processor groups PG1 to PGm. Are stored in the processor number storage areas 2-1 to 2-m corresponding to each processor group. Also, for example, based on the results of processor diagnosis performed at system initialization,
The number of processors that can execute a job among the processors included in each of the processor groups PG1 to PGm is stored in the available processor number storage areas 1-1 to 1-m corresponding to each processor group. After completing the above-described processing, the processor group management table initialization means 11 passes control to the processor group optimization means 12.

The table updating means 21 is notified from the firmware of the processor, for example, that the state of any of the processors has changed from the job executable state to the non-executable state due to a fault during the system operation after the system initialization. If, for example, the processor firmware notifies that the state has changed from the unexecutable state to the executable state due to the recovery of the failure, the available processor number storage area group 1 is updated accordingly. It performs a process as shown in FIG. That is, when the state of any of the processors changes during the operation of the system, the table updating unit 21 determines whether the change is from available to unusable or vice versa (step 301). In the case of a change from available to unavailable, the value of the available processor number storage area corresponding to the processor group to which the processor belongs in the available processor number storage area group 1 is decreased by 1 (step 302), and the processor group Control is passed to the optimization means 12 (step 304). Conversely, if the change is from unusable to usable, the value of the available processor number storage area corresponding to the processor group to which the processor belongs in the available processor number storage area group 1 is increased by 1 (step 303). Is passed to the processor group optimizing means 12 (step 304).

The processor group optimizing means 12 is a means for determining, for a processor group in which all processors in the group have become unusable, a processor group to substitute for execution of a job specified by the processor group, as shown in FIG. Such processing is performed. First, when control is passed from the processor group management table initialization means 11 or the table updating means 21, the available processor number storage area group 1 is referred to and the value of the available processor group storage area of all processor groups is 1 or more. It is checked (step 401) whether or not the value of all available processor number storage areas is not 1 or more, that is, if the value of at least one available processor group storage area is 0, the processor group is replaced. The process proceeds to step 402 to determine a processor group to be executed. Otherwise, the process skips step 402 and proceeds to step 403. In step 402, a processor group for executing the job specified by the processor group in place of the processor group is determined for all the processor groups for which the value of the available processor number storage area is 0, and the determined processor group number is set. Are stored in the execution processor group number storage area corresponding to the replacement source processor group in the execution processor group number storage area group 3. And step 40
Proceed to 3. At step 403, it is checked whether or not at least one job exists in the tightly coupled multiprocessor system. If there is, the process proceeds to step 404, where the control is passed to the job execution group changing unit 22, and the job execution group changing unit 22 does not exist. If it is, step 404 is skipped and the process ends.

In the determination of the alternative processor group in step 402 of the processor group optimizing means 12, the job designating the processor group having the available processor number storage area value of 0 is replaced with the job specifying the available processor number storage area value of 0. For all combinations in which processor groups that are not replaced are processed in units of processor groups, the load per usable processor of the processor group after the replacement, in which the value of the available processor number storage area is not 0, is entered by designating each processor group. The calculation is performed assuming that the number of jobs is substantially proportional to the number of processors stored in the number-of-processors storage area, and a processor group of the combination having the smallest load variation among all combinations is set as an alternative processor group. A specific example of this determination will be described later.

Next, the job execution group changing means 22
Is a means for changing a processor group to which a job existing in the system is to be executed, and performs processing as shown in FIG. When the control is passed from the processor group optimizing unit 12, the job execution group changing unit 22
A value of one available processor number storage area is input from the available processor number storage area group 1 to check whether the value is 0 (steps 501 and 502). If the value is 0, that is, If no processor is available, it is present in the system,
The job specified by the processor group is moved to the queue corresponding to the processor group of the number stored in the execution processor group number storage area corresponding to the processor group, and the execution destination is changed (step 50).
3). On the other hand, if the value is not 0, that is, if there is at least one available processor in the processor group, all the jobs specified by the processor group and assigned to the alternative processor group are deleted. The queue is moved to the queue corresponding to the specified processor group, and the execution destination is changed to the original processor group (504). The job execution group changing unit 22 repeats the above processing (501 to 504) for all processor groups (Step 505).

It is to be noted that the queue in the queue Q
The management may be performed by including the processor group designation in the job execution information stored in 1 to Qm, and the processor group designation and the actual assignment destination processor group are stored and managed in a table for each job. You may do it.

The job execution group selection means 31 is a means for selecting a processor group to which a job is to be submitted when a user submits a job designated by a processor group, and performs processing as shown in FIG. First, the job execution group selecting unit 31 refers to the value of the available processor number storage area corresponding to the processor group specified by the user in the available processor number storage area group 1 and determines whether the value is not 0. Is checked (step 601). If the value is not 0, there is at least one available processor, and the job input by the user is stored in the corresponding queue to be assigned to the specified processor group (step 602). On the other hand, if the value is 0, there is no usable processor in the specified processor group, and therefore the execution processor group number corresponding to the specified processor group in the execution processor group number storage area group 3 The data is stored in the corresponding queue to be assigned to the processor group of the number stored in the storage area (step 603).

Next, the operation of the present embodiment configured as described above will be described. In order to make the description easy to understand, as shown in FIG.
PG1, PG2, PG
3 is P11 and P12, P21 and P22, P31 and P32.
, And the processor group PG4 includes one processor P41.
Therefore, it is assumed that four areas exist in each of the available processor number storage area group 1, the processor number storage area group 2, and the execution processor group number storage area group 3 as shown in FIG.

At the time of system initialization, the processor group management table initialization means 11 stores all processor groups P
The number of processors included in G1 to PG4 is stored in the processor number storage areas 2-1 to 2-4. In this example, as shown in FIG. 7, “2” is stored in the area 2-1, “2” is stored in the area 2-2, “2” is stored in the area 2-3, and “1” is stored in the area 2-4. You. Next, each processor group PG1 to PG4
The number of available processors that can execute a job among the processors included in the storage area 1-1 is a usable processor number storage area 1-1.
To 1-4. Now, the processors P11, P12,
Assuming that the initial states of P21, P22, P31, P32, and P41 are available, unavailable, unavailable, available, available, available, unavailable, as shown in FIG. As described in FIG. 7, “1” is stored in the area 1-1, “1” is stored in the area 1-2, “2” is stored in the area 1-3, and “0” is stored in the area 1-4. After such processing,
The processor group management table initialization means 11 transfers control to the processor group optimization means 12.

In step 401 of FIG. 4, the processor group optimizing means 12 checks whether the values of the available processor number storage areas 1-1 to 1-4 of all processor groups are 1 or more. In the present example, the available processor group storage area 1 of the processor group PG4
Since the value of −4 is 0, the process proceeds to step 402 and performs the following processing.

The value of the available processor number storage area is 0
Is the processor group PG4
And there are the following three combinations of alternative destination processor groups. (1) The processor group PG1 is substituted for the processor group PG4. (2) The processor group PG4 is substituted for the processor group PG4. (3) The processor group PG3 is substituted for the processor group PG4.

Therefore, on the assumption that the number of submitted jobs specified by each processor group is substantially proportional to the number of processors stored in the number-of-processors storage area, the available processors in the alternative processor group are Calculate the load per unit.

(1) When the processor group PG1 is substituted for the processor group PG4, the load per available processor of the processor groups PG1, PG2, PG3 after substitution is calculated as follows. ○ Load of processor group PG1 = {(number of processors of PG1 = 2) + (number of processors of PG4 = 1)}
/ (Number of available processors of PG1 = 1) = 3 ○ load of processor group PG2 = (number of processors of PG2 = 2) / (number of available processors of PG2 =
1) = 2 Load of processor group PG3 = (number of processors of PG3 = 2) / (number of available processors of PG2 =
2) = 1

(2) When the processor group PG4 is replaced by the processor group PG2, the load per available processor of the replaced processor groups PG1, PG2, PG3 is calculated as follows. ○ Load of processor group PG1 = (number of processors of PG1 = 2) / (number of available processors of PG1 =
1) = 2 Load of processor group PG2 = {(number of processors of PG2 = 2) + (number of processors of PG4 = 1)}
/ (Number of available processors of PG2 = 1) = 3 ○ Load of processor group PG3 = (number of processors of PG3 = 2) / (number of available processors of PG2 =
2) = 1

(3) When the processor group PG3 is substituted for the processor group PG4, the load per available processor of the processor groups PG1, PG2, PG3 after substitution is calculated as follows. ○ Load of processor group PG1 = (number of processors of PG1 = 2) / (number of available processors of PG1 =
1) = 2 Load of processor group PG2 = (number of processors of PG2 = 2) / (number of available processors of PG2 =
1) = 2 Load of processor group PG3 = {(number of processors of PG3 = 2) + (number of processors of PG4 = 1)}
/ (Number of usable processors of PG2 = 2) = 1.5

If the case where the maximum value of the load is minimum is defined as the most equal, the maximum value of the load is 2 and the minimum value is 2 when the processor group PG3 of the above (3) is substituted. Means 12 determines that processor group PG4 is substituted for processor group PG4. Then, the determined number of the processor group PG3 is stored in the execution processor group number storage area 3-4 corresponding to the processor group PG4. Thereafter, the number of jobs in the system is determined. However, since the number is zero at the time of system initialization, the process ends.

Now, it is assumed that the user designates the processor group PG2 and submits the job A. The job execution group selecting means 31 determines in step 60 in FIG.
In step 1, the value of the available processor number storage area 1-2 corresponding to the processor group PG2 is referred to, and since the value is 1 and not 0, the job is assigned to the corresponding queue Q2 to be assigned to the specified processor group PG2. It is stored (step 602).

Next, at this point, the processor group PG
If the status of the processor P22 of the second processor P22 is changed from available to unavailable, the table updating means 21 determines in step 301 in FIG. 3 that the processor P22 has been disabled from available. 30
In 2, the value of the available processor number storage area 1-2 corresponding to the processor group PG2 to which the processor P22 belongs is changed from 1 to 0. Then, the control is passed to the processor group optimizing unit 12. In step 401 of FIG. 4, the processor group optimizing unit 12 checks whether the values of the available processor group storage areas 1-1 to 1-4 of all the processor groups are 1 or more. In the present example, the processor groups PG2, PG4
Available processor group storage areas 1-2, 1-4
Is 0, the process proceeds to step 402, and the following processing is performed.

The value of the available processor number storage area is 0
Is the processor group PG
2 and PG4. There are the following four combinations of alternative destination processor groups. (1) The processor group PG1 replaces the processor group PG2, and the processor group PG3 replaces the processor group PG4. (2) The processor group PG1 replaces the processor group PG4, and the processor group PG3 replaces the processor group PG1. (3) The processor group PG1 is substituted for the processor groups PG2 and PG4. (4) The processor groups PG2 and PG4 are substituted for the processor group PG3.

Therefore, on the assumption that the number of submitted jobs specified by each processor group is substantially proportional to the number of processors stored in the number-of-processors storage area, the available processors in the replacement destination processor group are Calculate the load per unit.

(1) When the processor group PG2 is replaced by the processor group PG1 and the processor group PG4 is replaced by the processor group PG3, the load per one available processor of the replaced processor groups PG1 and PG3 is calculated as follows. become that way. ○ Load of processor group PG1 = {(number of processors of PG1 = 2) + (number of processors of PG2 = 2)}
/ (Number of available processors of PG1 = 1) = 4 ○ Load of processor group PG3 = {(number of processors of PG3 = 2) + (number of processors of PG4 = 1)}
/ (Number of usable processors of PG3 = 2) = 1.5

(2) When the processor group PG1 is replaced with the processor group PG4 and the processor group PG3 is replaced with the processor group PG2, the load per one available processor of the replaced processor groups PG1 and PG3 is calculated as follows. become that way. ○ Load of processor group PG1 = {(number of processors of PG1 = 2) + (number of processors of PG4 = 1)}
/ (Number of available processors of PG1 = 1) = 3 ○ Load of processor group PG3 = {(number of processors of PG3 = 2) + (number of processors of PG2 = 2)}
/ (Number of usable processors of PG3 = 2) = 2

(3) When the processor groups PG2 and PG4 are replaced by the processor group PG1, the load per available processor of the replaced processor groups PG1 and PG3 is calculated as follows. ○ Load of processor group PG1 = ｛(number of processors of PG1 = 2) + (number of processors of PG2 = 2) +
(Number of processors of PG4 = 1) １ / (Number of available processors of PG1 = 1) = 5 ○ Load of processor group PG3 = (Number of processors of PG3 = 2) / (Number of available processors of PG3 =
2) = 1

(4) When the processor group PG3 is replaced with the processor group PG2, PG4, the load per usable processor of the replaced processor group PG1, PG3 is calculated as follows. ○ Load of processor group PG1 = (number of processors of PG1 = 2) / (number of usable processors of PG3 =
1) = 2 Load of processor group PG3 = ｛(number of processors of PG3 = 2) + (number of processors of PG2 = 2) +
(Number of processors of PG4 = 1) / (Number of usable processors of PG3 = 2) = 2.5

Here, if the case where the maximum value of the load is the smallest is defined as the most equal, the maximum value of the load is 2.5 when the processor group PG2 and PG4 are replaced by the processor group PG3 of (4). Since it is the smallest, the processor group optimizing means 12 sets the processor group PG
3 is determined to replace the processor groups PG2 and PG4. The number of the processor group PG3 is stored in the execution processor group number storage area 3-2 corresponding to the processor group PG2.
Execution processor group number storage area 3-4 corresponding to
Of the processor group PG3. Thereafter, the number of jobs in the system is determined, and since it is not 0, the control is passed to the job execution group changing unit 22.

The job execution group changing means 22 is provided in FIG.
In steps 501 and 502, the available processor number storage area 1 corresponding to the processor group PG2
Recognizing that the value of -2 is 0, in step 503, the number of the processor group PG3 stored in the execution processor group number storage area 3-2 corresponding to the processor group PG2 is input and assigned to the processor group PG2. In order to execute the executed job A in the processor group PG3, the job A in the queue Q2 corresponding to the processor group PG2 is
Move to queue Q3 corresponding to G3. As a result, the job A specified by the disabled processor group PG2 is executed in the processor group PG3.

Although not specifically illustrated above, if at least one processor of the unusable processor group PG2 or PG4 becomes available, the processor group PG2 or PG4
When the job designated by 4 remains in the queue of the alternative processor, the job is changed to be executed in the original processor groups PG2 and PG4.

Table updating means 2 in the above embodiment
1 transfers control to the processor group optimizing unit 12 every time the value of the available processor number storage area 1-1 to 1-m is updated as shown in FIG. 3, for example, as shown in FIG. Steps 305 and 306 are added, and the control is transferred to the processor group optimizing means 12 only when the value of the available processor number storage area 1-1 to 1-m becomes 0 when the value is decreased by 1. Alternatively, the control may be passed to the processor group optimizing means 12 only when the value of the available processor number storage areas 1-1 to 1-m becomes 1 when the value is increased by 1. In this case,
The replacement processor group is not reviewed when the number of available processors in the replacement processor group or the like increases or decreases. However, due to an increase or decrease in the number of available processors, a new processor group in which all processors become unusable appears, or conversely, one processor group in which all processors are unusable becomes available. At this point, the review will take place.

[0056]

As described above, according to the present invention, the following effects can be obtained.

In a tightly-coupled multiprocessor system in which a plurality of processors are divided into several processor groups to execute a job by designating a processor group, one or more processor groups are in a state where the job cannot be executed due to a processor failure. In this case, jobs are distributed according to the number of processors available at that time and the original processor group configuration, so that a system operation with high throughput can be performed without concentrating a load on a specific processor group.

In general, the load on each processor group of a job input by a user is taken into consideration in consideration of the fact that all processors in the system can be used, and the number of processors belonging to each processor group and the Since the alternative destination processor group is determined only by simple information such as the number of processors that can execute the job, the configuration is simplified.

Since the replacement destination processor group is appropriately reviewed, more appropriate replacement processing can be performed. In particular, in a configuration in which the table updating means transfers control to the processor group optimizing means every time the available processor number storage area is updated, a change in the processor state in the alternative processor group or the like (change from available to unusable or vice versa) Since the alternative processor group is reviewed at the time of (1), more appropriate alternative processing can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a tightly coupled multiprocessor system to which the present invention is applied.

FIG. 2 is a block diagram of an embodiment of an alternative processing method in the tightly coupled multiprocessor system of the present invention.

FIG. 3 is a flowchart illustrating a processing example of a table updating unit.

FIG. 4 is a flowchart illustrating a processing example of a processor group optimizing unit;

FIG. 5 is a flowchart illustrating a processing example of a job execution group changing unit;

FIG. 6 is a flowchart illustrating a processing example of a job execution group selection unit;

FIG. 7 is a diagram showing a specific example of the number of processors included in a processor group, and the number of areas included in an available processor number storage area group, a processor number storage area group, and an execution processor group number storage area group;

FIG. 8 is a flowchart illustrating a processing example of a table updating unit in another embodiment of the alternative processing method in the tightly-coupled multiprocessor system of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Available processor number storage area group 1-1-1-m ... Available processor number storage area 2 ... Processor number storage area group 2-1-2-m ... Processor number storage area 3: Execution processor group number storage area Group 3-1 to 3-m execution processor group number storage area 11 processor group management table initialization means 12 processor group optimization means 21 table update means 22 job execution group change means 31 job execution group selection means

Claims (3)

(57) [Claims] In a tightly coupled multiprocessor system in which a plurality of processors sharing a main storage device are divided into several processor groups and a job is executed by designating each processor group, each processor group has a one-to-one correspondence. A number-of-processors storage area for storing the number of processors belonging to the processor group, and a job execution area provided for each processor group in a one-to-one correspondence, among the processors belonging to the processor group. An available processor number storage area for storing the number of possible processors, and an execution processor provided in one-to-one correspondence with each processor group and storing the number of another processor group that substitutes for the job specified by the processor group Group number storage area and system At initialization, the number of processors included in each processor group and the number of available processors are stored in the processor number storage area and available processor number storage area corresponding to each processor group. Processor group management table initialization means for transferring control to the conversion means, and the number of available processors corresponding to the processor group to which the processor belongs when a processor that cannot execute a job occurs during system operation When the value of the storage area is decremented by 1, and control is passed to the following processor group optimizing means, and a processor capable of executing the job occurs, the number of available processors corresponding to the processor group to which the processor belongs Increase the value of the storage area by 1 A table updating unit that transfers control to the following processor group optimizing unit; and a unit for which the value of the available processor number storage area is 0 every time control is passed from the processor group management table initializing unit or the table updating unit. For the processor group of (1), the number of the processor group which substitutes the job specified by the processor group is determined and stored in the execution processor group number storage area corresponding to the substitution source processor group. A processor group optimizing means for transferring control to the job execution group changing means of the job execution group changing means; and a processor group for which the value of the available processor number storage area is not 0 for all the processor groups. In the case of a processor group in which the value of the available processor number storage area is 0, the job assigned to the subgroup is assigned to the specified processor group, and the job specified by the processor group is stored in the execution processor group number corresponding to the processor group. A job execution group changing unit to be assigned to another processor group corresponding to the number stored in the area, and when a job specified by the processor group is submitted, the value of the available processor number storage area corresponding to the specified processor group is not 0 Is assigned to the specified processor group, and if the value of the available processor number storage area corresponding to the specified processor group is 0, the job is stored in the execution processor group number storage area corresponding to the specified processor group. And a job execution group selecting unit for allocating a job to a processor group of a given number, and wherein the processor group optimizing unit can use a job specifying a processor group having a value of 0 in the number of available processors storage area. For all combinations in which a processor group having a non-zero value in the number-of-processors storage area is substituted for each processor group, the load per available processor of the processor group having a non-zero number of usable-processor-numbers storage area after the substitution is performed. Is calculated on the assumption that the number of input jobs specified by each processor group is substantially proportional to the number of processors stored in the processor number storage area, and the load is applied to the combination with the smallest load variation among all the combinations. Processor group Alternate processing method in tightly coupled multiprocessor system and having a configuration for determining a replacement processor group. 2. A processor corresponding to a processor group to which a processor that cannot execute a job occurs during system operation, instead of the table updating unit according to claim 1. When the value of the processor number storage area is reduced by 1 and the value becomes 0, the control is transferred to the processor group optimizing means, and if a processor capable of executing a job occurs, the processor to which the processor belongs A tightly coupled multiprocessor system characterized by comprising a table updating means for increasing the value of the available processor number storage area corresponding to the group by 1 and transferring the control to the processor group optimizing means when the value becomes 1. Alternative processing method. 3. The processor group optimizing means includes:
For all combinations in which a job that specifies a processor group whose available processor number storage area value is 0 is replaced by a processor group whose available processor number storage area value is not 0, the available processor number storage after replacement is stored. For the processor group whose area value is not 0, the value of the available processor number storage area corresponding to the relevant processor group and the value of the available processor number storage area corresponding to the relevant processor group and the processor group to be substituted by the relevant processor group And calculating the ratio of the sum with the value of the number-of-processors storage area corresponding to the combination, and determining the combination having the minimum value of the ratio among all the combinations as the combination having the smallest load variation. Item 7. A tightly coupled multiplex according to item 1 or 2. Alternate processing method in processor system.