JPH03113563A - Multiprocessor scheduling method - Google Patents

Abstract

PURPOSE:To improve the performance of cache memory and to reduce the load of a bus by controlling the processes turned into groups and carrying out the processes in the same group via the same processor. CONSTITUTION:When the processes having a large quantity of shared data are carried out by different processors 13-1 - 13-3, the communication is frequently produced to secure the coincidence of caches. As a result, the cache performance is deteriorated. In this respect, such processes 12-1 - 12-8 are turned into the groups 15-1 - 15-3 and carried out by the same processors 13-1 - 13-3. Thus no communication is required for coincidence of caches. Then the groups are carried out by a specific one of processors 13-1 - 13-3 and therefore the processes which are left unexecuted for a fixed time or longer can be carried out by any of processors 13-1 - 13-3 so that the load is not biased among the processors 13-1 - 13-3. As a result, the performance of a cache memory is improved and the load can be scattered.

Description

[Detailed Description of the Invention] [Industrial Application Field] In a multiprocessor operating system that runs multiple processes in parallel in a multiprocessor system consisting of multiple processors, this is particularly useful when constructing a high-speed system or when using a single processor. The present invention relates to a process scheduling method suitable for operating a program created for a computer without any changes on a multiprocessor system.

[Prior Art] Multimer δ-microprocessors using multiple microprocessors are being actively developed. Japanese Patent Publication No. 56-1
No. 27261, a multiprocessor system, is an example of this, and shows a system configuration in which a plurality of processors are connected by paths. Operating systems for multiprocessors are also being actively developed.

Unlike single-processor systems, multiprocessor operating systems need to be able to run multiple processes on different processors at the same time, and therefore need to have a way to decide which processor to assign a process to. There is. In this case, the most common strategy is to allocate the processor to the processor with the least load in order to maximize processor utilization efficiency. This strategy is generally easy to implement and is therefore widely adopted. This is because the process scheduler of the operating system running on each processor can select the next process to be executed in accordance with the priority of the process, as in the case of a single processor system. On a heavily loaded processor, the process takes a long time to execute, and the process scheduler is activated less frequently.

Conversely, a processor with a light load often enters an idle state with no processes to process, and therefore there are many times when the process scheduler is activated to determine whether to transition to an idle state. Therefore, processors with light loads are given more opportunities for process scheduling, and the loads among the processors are automatically balanced. The above strategy is a reasonable process allocation strategy in a homogeneous multiprocessor system with the same functions.

Also, for information on how to manage and schedule processes as a group, see the Experimental Using Multiprocessor Systems Zoom Status Report J.
Using Multi-processor
Systems-A 5tatus Report
(ACM Computing Survey
sy Vol. 12, No. 2, p. 121-165.1980)
g). This allows processes that can be processed in parallel to be grouped and scheduled to be executed simultaneously on different processors as much as possible.

[Problems to be Solved by the Invention] A process scheduler that has a function of grouping a plurality of processes and executing the processes belonging to the same group on the same processor as much as possible is preferable.

The processes are grouped here because of the following user requirements and requirements for constructing hardware and operating systems (note that this group is a collection of processes that handle the same data).

In a multiprocessor using a shared bus and shared memory,
Each processor has a cache memory to reduce the load on the bus. In such a system, in order to ensure data consistency between caches, when one processor updates data on the cache, the updated result is transferred to the cache that has the same data. If processes that share a lot of data are executed on different processors, the number of data transfers between caches will increase, and the performance of the cache will not be maximized.

Increased bus load: In conventional multiprocessor systems, sufficient consideration has not always been given to this problem.

Furthermore, it is desirable that existing software developed for a single processor be executed on a multiprocessor system without any changes. In a multiprogramming environment, contention among multiple processes can occur, as is well known.

To solve this problem, 1. Normally, when accessing shared data, it is necessary to perform exclusive control using a semaphore mechanism, etc., but the processing overhead of the semaphore mechanism becomes a problem, so it is necessary to In many cases, measures are taken to disable interrupts by raising the processor interrupt mask, which has a shorter processing time than the above mechanism. These means utilize the fact that when viewed microscopically on a single processor, processing on the processor is always performed sequentially.

In reality, there are many software programs that do not perform exclusive control over shared data for performance reasons.

For this reason, there is a problem in that when existing software is executed on a multiprocessor system, normal operation is not necessarily guaranteed.

Conventional group scheduling is suitable for parallel processing because it operates a group of processors on different processors, but it cannot deal with the problems described above.

The earlier application related to this application is a patent application filed by the same applicant.
~60092 "Operating System for Multiprocessors". This prior application has a process management table in which the processor share occupied by a process is recorded, and processes under the control of other processors are intercepted under certain conditions and made to perform processing under its own management. In this prior application, there is no record of the last used processor or description of process management for each group.

An object of the present invention is to: In a multiprocessor system,
An object of the present invention is to provide a process scheduling method that can improve the performance of a cache memory and reduce the load on a bus by managing processes in groups and executing the processes in the same group with the same processor.

Another object of the present invention is to provide a process scheduling method that allows existing software for a single processor system to be used as is without any modification.

Still another object of the present invention is to provide a process scheduling method that can maintain an equal workload among processors even when a group of processes is executed by the same processor.

[Means for Solving the Problems] In order to achieve the above object, a process scheduling method according to the present invention provides a multiprocessor system including a plurality of processors, in which the plurality of processors are controlled by a single operating system. In the process scheduling method, the process scheduler of the operating system allocates a processor to a process according to a process management table for managing processes, and describes a group identifier that allows duplication in the process management table, and assigns a group identifier to the group. A group management table is provided to store information on processes having the same identifier, and processor allocation information for instructing a processor to execute a process in the group is written in the group management table. For the selected process, the group of the process is determined by referring to the process management table, and the process is operated on the processor specified by the processor allocation information of the group management table of the group, so that the same group Processes belonging to the same processor are executed by the same processor.

The process scheduler stores the execution waiting time of the process in the process management table, refers to the process management table for the process selected to be executed next, and when the execution waiting time is equal to or greater than a certain value,
When the execution wait time is less than or equal to a certain value when the process is run on an arbitrary processor, the process management table is referred to, the group of the process is determined, and the processor specified by the processor allocation information in the group management table of the group is executed. The process may be run on top of the computer.

Furthermore, the time during which no process in the group is being executed, that is, the group execution waiting time, is stored in the group management table, and the process scheduler refers to the process management table for the process selected to be executed next. Then, the group of the process may be determined, and when the group execution waiting time in the group management table of the group is equal to or greater than a certain value, the processor allocation information in the group management table may be changed.

Furthermore, the execution waiting time of the process and the identifier of the processor on which the process was last executed are stored in the process management table, and the process scheduler refers to the process management table for the process selected to be executed next. So, when the waiting time is less than a certain value,
The process may be executed by the processor that executed it last time or by the processor specified by the processor allocation information in the group management table, and when the value exceeds a certain value, it may be possible to operate by a processor other than the above.

Furthermore, instead of the execution waiting time of the process, the number of times the process is refused to be assigned to a processor may be used.

Furthermore, instead of the time during which no process in the group is being executed, that is, the group execution waiting time, we used the number of times the process was refused to be assigned to a processor.

As another process scheduling method of the present invention, in a multiprocessor system that includes a plurality of processors and is controlled by a single operating system, a processor that needs to switch the process being executed , a process scheduler that executes the process scheduler of the operating system according to a process management table that manages processes, and determines the process that the processor will execute next.
The scheduling method describes group identifiers that allow duplication in the process management table, provides a group management table that stores information on processes with the same group identifier, and stores processes within the group in the group management table. Processor allocation information instructing a processor to execute is written, and the process scheduler executes the process according to the processor allocation information in the group management table, which is specified by the group identifier in the process management table, from among executable processes. The process indicated by the currently running processor may be executed next.

The execution waiting time of the process is stored in the process management table, and the process scheduler refers to the process management table when selecting a process to be executed next, and when the execution waiting time is equal to or greater than a certain value, the process scheduler selects the process to be executed next. The process may be made executable.

Furthermore, the time during which no process in the group is executed, that is, the group execution waiting time, is stored in the group management table, and when the process scheduler selects the next process to be executed, the process management table is stored. reference to determine the group of processes in question;
When the group execution waiting time in the group management table for the group is greater than or equal to a certain value, the processor allocation information in the group management table may be changed to enable the process to be executed.

Furthermore, the process management table stores the execution waiting time of the process and the identifier of the processor on which the process was last executed, and the process scheduler refers to the process management table when selecting the process to be executed first. When the waiting time is less than a certain value, the processor that previously executed the process or the processor indicated by the processor allocation information in the group management table is the same as the processor that is executing the process scheduler. It may be set to be executable only when the waiting time is greater than or equal to a certain value.

Furthermore, instead of the execution waiting time of the process, the number of times the process is refused to be assigned to a processor may be used.

Furthermore, instead of the time during which no process in the group is being executed, that is, the group execution waiting time, the number of times the process was refused to be assigned to a processor was used.

As another process scheduling method of the present invention, in a multi-processor system including a plurality of processors, the plurality of processors are controlled by a single operating system. A process scheduling method in which a process scheduler assigns any processor zO- to a process,
Describe the priority and group identifiers that allow duplication for each of the plurality of processors in the process management table,
A group management table is provided for storing information on the processes having the same group identifier, and processor allocation information for instructing a processor to execute a process in the group is provided in the group management table, and the group identifier of the process management table indicates The priority level corresponding to the processor indicated by the processor allocation information in the group management table may be set high.

[Operation] In a multiprocessor system consisting of a plurality of processors, the process scheduling method of the present invention intentionally groups processes that can be executed by any of the plurality of processors and assigns a specific processor to execute the group. It is intended to be limited.

When processes that have a lot of shared data are executed by different processors, communication for cache matching occurs frequently, reducing cache performance.

In the present invention, such processes can be grouped and executed by the same processor, thereby eliminating the need for communication for cache matching and improving performance. Furthermore, in order to prevent the load from becoming unbalanced among processors by having a group executed by a specific processor, processes that have not been executed for a certain period of time can be executed by any processor. Load balancing is possible. However, this feature can also be disabled by setting a process fixation flag.

Furthermore, the present invention makes it possible to operate a group of processes that require exclusive control on the same processor,
Normal operation can be guaranteed even if an existing process for a conventional single processor that does not consider exclusive control over shared memory is used without modification. This does not mean that other processors are not used, but for processes other than those in the process group mentioned above, the multiprocessor function is assumed to be executable on any processor without such restrictions. can be used effectively.

[Example] An example of the present invention will be described below with reference to FIG. In FIG. 1, 13-1 to 13-2 are processors, 12
-1 to 12-8 are processes, and 15-1 to 15-3 are groups.

Processors 13-1 to 13-3 process processes 12-1 to 13-3.
Execute while switching 12-8. Process 12-1
~12-3 is the process 12-3 as the group 15-1.
4 and 12-5 as group 15-2, process 12
-6 to 12-8 are grouped together as group 15-3. Instructions on how to group processes are provided by the user, but may also be provided automatically by the compiler or operating system.

The processors that execute processes are determined for each group, and processes belonging to the same group are executed by the same processor. Although the number of groups and the number of processors are equal in FIG. 1, these numbers do not have to match, and either one may be larger or smaller.

FIG. 2 shows a system configuration according to the present invention. In FIG. 2, 13-1 to 13-3 are processors, 11-1 to 1
1 to 3 are processor numbers, 12-1 to 12-6 are processes, 14-1 is a timer, 4 is a connection bus, 5 is a shared memory, 31 is an OS program, 311 is a scheduler, 3
2-1 to 32-4 are process management tables, 321 to 1
~321~4 are processor identifiers, 322-1~322
-4 is the priority, 323-1 to 323-4 are execution waiting times, and 324-1 to 324-4 are counters.

325-1 to 325-4 are process fixed flags, 326
-1 to 326-4 are group identifiers, 33-1 to 33-
2 is a group management table, 331-1 to 331-2 are processor identifiers, 332-1 to 332-2 are execution waiting times, 333-1 to 333-2 are counters, 334-1
~334-2 is a group fixed flag.

The processors 13-1 to 13-3 are connected to each other and to the shared memory 5 by a connection bus 4. Each processor has a processor number 11-1-11-3, respectively. In this embodiment, processors 13-1 to 13
-3 have processor numbers "1", "2", and "3", respectively. In this embodiment, the processor number is set using a hardware switch, but other methods include storing it in advance in a hardware ROM, or setting it using a program at startup.

One of the processors 13-1 to 13-3 has a timer 14-1, and a timer interrupt enters that processor at regular intervals.

Each process runs on a processor. shared memory 5
includes an operating system program 31 and a process scheduler 3 that is part of this program 31.
11. Process management table 32 (32-1), which is a part of the tables managed by this operating system.
~32-3), group management table 33 (33-1
~33-2) exists.

The following information is placed in the process management table 32. First, a processor identifier 21 (321
1-321-4). In the embodiment, the processor identifier of the process management table 32-1 is "1", and the process 0 (1) corresponding to this process management table is "1".
2-1) has "1" in processor numbers 11 to 1,
It shows what should be executed by the processor 13-1. If the processor that will execute the process is undecided, set this processor identifier to O. If you do not decide which processor will execute the process and want to be able to execute the process on any processor, set the processor identifier to a negative value. Keep it.

Second, the priority level 322 (
322-1 to 322-4). The smaller the priority value, the higher the priority of the process, indicating that it must be executed in priority over other processes.

The third is the execution waiting time 323 (323-1 to 323-
4). This execution waiting time indicates the time period during which the process is in the execution waiting state. When the process is executed, this execution waiting time is returned to 0.

The fourth is the counter 324 (324-1 to 324-4)
It is. The counter is a counter that counts the number of times execution of a process is refused.

The fifth is the process fixed flag 325 (325-1 to 3
25-4). This contains information on whether the process is fixed to a processor. If it is fixed, the process will not be executed on any processor other than the one specified in the processor identifier 321.

The sixth is the group identifier 326 (326-1 to 326
-4). This contains information indicating which group the process belongs to. There is a one-to-one correspondence between the group identifier and the group management table.

The group management table contains the following information. First, there is the processor identifier.

This runs processes belonging to that group -Z'
/ - The processor number is written.

The second is the execution waiting time 322 (322-1 to 322-
2). This execution waiting time is the time during which none of the processes belonging to the group is being executed by the processor and one of them is in the execution waiting state. When even one process in the group is executed, this execution waiting time is returned to O.

The third is a counter. This counter counts the number of times a process belonging to the group is refused execution. It is cleared to O when a process in the group is executed.

The fourth is a group fixation flag, and when this flag is set, the processor that executes the group is fixed.

FIG. 3 shows the operating system program 31.
The algorithm of the process scheduler 311, which is part of the process scheduler 311, is shown.

First, from among the processes waiting to be executed, the process to be executed next is selected according to the priority order (31
1-1) Read the group identifier 326 from the process management table of the selected process. The group management table is searched based on this group identifier (311-2).

Processor identifier 331 in this group management table
is read and the following processing is performed based on this value (311
~3).

If the value is a positive value, the processor executing the scheduler program 311 reads its own processor number from 11 (11 to 1 to 1l-3) and combines it with the value of the positive processor identifier 331. Compare (311
~5), if they do not match, select the process with the next highest priority (311-1); if they match, assign the selected process to your processor (311-1);
311-6).

When the processor identifier in the group management table is 0, the value of the processor identifier is changed to the value of the own processor number (11) (311-4), and the selected process is assigned to the own processor (311-6).
).

If the processor identifier is a negative value, the selected process is unconditionally assigned to the own processor (31
1-6).

Figure 4 shows program 3 of the orting system.
1 shows another embodiment of the algorithm of the process scheduler 311, which is part of the process scheduler 311 of FIG.

First, from among the processes waiting to be executed, the process to be executed next is selected according to the priority order (31
1-1), Process management table 3 of the selected process
2, the processor identifier 321 is read out and compared with its own processor number 11 (311-7). If the values are equal, the selected process is assigned to the own processor (311-6).

If they are not equal, it is determined whether the process fixed flag 325 is set (311-8). If it is, the process with the next highest priority is reselected (3
11-1).

If not, check whether the execution waiting time 323 of the process management table 32 exceeds the threshold (
311-9).

If it exceeds the number, set the processor identifier 321 of the process management table 32 to the own processor number (3
11-14), Assign the selected process to yourself (
311-6).

If the number has not been exceeded, a group management variable is searched from the group identifier 326 in the process management table 32, and a check is made to see if the processor identifier therein is equal to the own processor number (311-10).

If they are equal, steps 311 to 14 are executed again.

If they are not equal, the group fixed flag 334 in the group management table 33 is checked (311-11).

When the flag is set, the process with the next highest priority is checked (311-1).

When the flag is not set, it is checked whether the execution waiting time 332 of the group management table 33 exceeds a threshold value. If the priority has not been exceeded, the process with the next highest priority is checked (311-1).

If it exceeds the number, set the processor identifier 331 of the group management table 33 and the processor identifier 321 of the processor management table 32 to its own processor number 11 (311-13, 311-14), and execute the selected process. (311-6).

FIG. 5 shows the operating system program 31.
3 shows another embodiment of the algorithm of the process scheduler 311 that is part of the process scheduler 311 .

First, from among the processes waiting to be executed, the process to be executed next is selected according to the priority order (31
1-1). Process management table 3 of selected process
2, the processor identifier 321 is read out and compared with its own processor number 11 (311-7). If the values are equal, the selected process is assigned to its own processor (311-6).

If they are not equal, it is determined whether the process fixed flag 325 is set (311-8). If it is, the counter 324 of the process management table 32 and the counter 333 of the group management table 33 are incremented (311-17, 311-18), and the process with the next highest priority is reselected (311-18). 1).

When the process fixed flag is not set, it is checked whether the counter 325 of the process management table 32 exceeds the threshold value (311-15).

If it exceeds the number, set the processor identifier 321 of the process management table 32 to the own processor number (3
11-14), Assign the selected process to yourself (
311-6).

If the number has not been exceeded, the group identifier 326 in the process management table 32 is searched for in the proof management table, and a check is made to see if the processor identifier therein is equal to the own processor number (311-10).

If they are equal, steps 311 to 14 are executed again.

If they are not equal, the group fixed flag 334 in the group management table 33 is checked (311-11).

When the flag is set, steps 311 to 17 are executed.

When the flag is not set, it is checked whether the counter 333 of the group management table 33 exceeds the threshold value. If not, execute steps 311 to 17.

If it exceeds the number, set the processor identifier 331 of the group management table 33 and the processor identifier 321 of the processor management table 32 to its own processor number 11 (311-13, 311-14), and execute the selected process. (311-6).

FIG. 6 describes in detail the method of allocating the selected processes 311 to 6 in FIGS. 3, 4, and 5 to the own processor.

First, the registers of the processes that have been executed so far are saved (312-1). This is done so that when the process returns to execution later, it can resume from where it left off.

Next, the execution waiting time 323 of the process management table 32 of the selected process is set to O (312-2).

Similarly, the execution waiting time 332 of the group management table 33 of the group to which the selected process belongs, the counter 324 of the process management table 32. The counter 333 of the group management table is also set to 0 (respectively, 312-3
, 312-4, 312-5).

Since each process has an independent user space, the user space is switched to that of the selected process (312-6).

Finally, the values of the registers saved when the process was interrupted last time are restored (312-7). Since the saved registers include the program counter, the process is restarted from the place where it was interrupted last time.

Reference numeral 313 in FIG. 7 shows an interrupt processing routine sent to the processor by the timer in FIG. 14-1. Timer interrupts occur every 10 seconds.

When a timer interrupt occurs, the processor first checks all the process management tables 32, and for processes that are waiting for execution, their execution waiting time 32 is checked.
3 is incremented (313-1).

Next, the entire group management table 33 is examined, and the waiting time 333 of the group that does not include any running processes but includes processes waiting to be executed is incremented (313-
2).

Another embodiment of the system configuration diagram of the present invention shown in FIG. 2 is shown in FIG. In FIG. 8, 13-1 to 13-3 are processors, 11-1 to 11-3 are processor numbers, and 12-1
~12-6 is a process, 14-2 is a timer, 4 is a connection bus, 5 is a shared memory, 31 is an OS program, 311
is a scheduler, 32-1 to 32-4 are process management tables, 321-1 to 321-4 are processor identifiers,
325-1 to 325-4 are process fixed flags, 326
-1 to 326-4 are group identifiers, 327-1 to 32
7-4 is CPU time, 328-1 to 328-4.329
-1 to 329-4, 330-1 to 330-4 are priorities, 33-1 to 33-2 are group management tables, 331
~1~331'-2 are processor identifiers, 334-1~
334-4 is a group fixed flag.

The processors 13-1 to 13-3 are connected to each other and to the shared memory 5 by a connection bus 4. Each processor has a processor number 11-1-11-3, respectively. In this example, Procera f13-1 to 13
-3 have processor numbers "1", "2", and "3", respectively.

One of the processors 13-1 to 13-3 has a timer 14-2, and a timer interrupt is sent to that processor at regular intervals.

In the embodiment, two types of interrupts are generated: every 10 m5 and every 18 m.

Each process runs on a processor. shared memory 5
includes an operating system program 31 and a process scheduler 3 that is part of this program 31.
14. Process management table 32 (32-1
~32-3), group management table 33 (33-1
~33-2) exists.

The following information is placed in the process management table 32. First, a processor identifier 21 (321
1-321-4). In the embodiment, the processor identifier of the process management table 32-1 is "1", and the process 0 (1) corresponding to this process management table is "1".
2-1) has "1" in processor numbers 11 to 1,
It shows what should be executed by the processor 13-1. When the processor that executes the process is undetermined, this processor identifier is set to 0. Furthermore, if you do not specify a specific processor to execute the process and want to be able to execute the process on any processor, set the processor identifier to a negative value.

The second is Plow ↑ 2 fixed flag 325 (325-1
~325-4). This contains information on whether the process is fixed to a processor. If it is fixed, the process is written in the processor identifier 321.

The third is the group identifier 326 (326-1 to 326
-4). This contains information indicating which group the process belongs to. There is a one-to-one correspondence between the group identifier and the group management table.

The fourth is CPU time. This is the CPU time used by the process and is used in priority calculations. However, this CPU time is not accurate because it is divided by 1/4 every second.

Fifth, priorities 328 to 3 determine the execution order of processes.
It is 30. Each process has multiple priorities because each processor has its own priority. The smaller the priority value, the higher the priority of the process, indicating that it must be executed in preference to other processes with higher priority values.

In the group management table, the processor identifier 331
(331-1-331-2). This indicates the processor number that executes the process belonging to that group.

FIG. 9 shows the scheduler program 31 in FIG.
4 algorithms are shown. A processor running a scheduler program sequentially checks the priorities in the process management table that correspond to it. Then, the process with the highest priority (the lowest priority number) is selected (314-1). However, at this time, processes for which the process fixed flag 325 of the process management table 32 is set and whose process identifier 321 is not its own processor number are excluded.

Next, the selected process is assigned to its own processor (314-2).

315 in FIG. 10 is 10 generated by the timer 14-2.
=39- This is an interrupt processing routine for each Oboro S. Here, each processor increments the CPU time 327 in the process management table 32 of the process being executed.

316 in FIG. 11 is the ls generated by the timer 14-2.
This is the interrupt processing routine for each. Here, the priorities 328 to 330 of each process management table 32 are recalculated. 316-1 processes each process, and 316-3 processes each processor. In 316-2, the CPU time 327 is set to 1/4 every second to prevent it from becoming too large.

The priority is calculated in 316-4 to 316-7. The priority is basically equal to the CPU time of the process, but the priority is the processor identifier 321 of the process management table 32.
Add 20 to the priorities of processors not listed in . Also, the priority of the processor that is not written in the processor identifier of the group management table 33 is still 20.
Add.

As a result, the priority of the process becomes higher for the processor written in the processor identifier of the process management table or the group 40-management table, and the process is more likely to be executed.

FIG. 12 shows a case where a program developed for a single processor is executed on a multiprocessor using a conventional scheduling method. In FIG. 12, 12-
1 to 12-2 are processes, 41 to 1 to 41 to 2 are interrupt disabling processes, 42-1 to 42-2 are resource usage processes, 43
-1 to 43-2 are interrupt permission processing.

A single processor has only one process running at any moment. Therefore, many programs that are intended to be executed on a single processor simply disable interrupts even when using shared resources.

This is because processing is faster than locking and using resources. With a single processor, if you disable interrupts, only one process is running at the moment, so resources are not used by multiple processes at the same time.

However, in a multiprocessor, as shown in Figure 12,
Even if interrupts are disabled in 41-1 and 41-2, 42-
A plurality of processes such as 1 and 42-2 may use resources at the same time.

FIG. 13 shows a case where a single processor program is executed by a multiprocessor using the scheduling method of the present invention. If processes 12-1 and 12-2 that use the same resources are placed in the same group and executed on processor 0, the resource usage of 42-1 and 42-2 will always be executed serially, and the resource usage will be reduced. There is no competition.

In addition, in recent years, instead of the traditional process that has individual programs and data, threads and lightweight processes that share programs and data have been introduced.
light weight process
) is attracting attention.

Threads that share programs and data share a large amount of data, so the improvement in cache performance when grouped is even greater.

Furthermore, if threads that share programs and data are grouped together and executed on a single processor, there is often no need to switch between programs and data.
The O8 processing time can be shortened.

Therefore, the present invention is highly effective even when applied to such threads and lightweight processes.

[Effects of the Invention] In a multiprocessor system consisting of a plurality of processors, the process scheduling method of the present invention intentionally groups processes that can be executed by any of the plurality of processors and identifies a processor to execute the group. It is intended to be limited to the following processors.

When processes with a large amount of shared data are executed by different processors, communication for cache number ratio occurs frequently, reducing cache performance.

In the present invention, since such processes can be grouped and executed by the same processor, communication for cache 44-coherence is not necessary, and performance can be improved. Furthermore, in order to prevent the load from becoming unbalanced among processors by having a group executed by a specific processor, processes that have not been executed for a certain period of time can be executed by any processor. Load balancing is possible. However, this feature can also be disabled by setting a process fixation flag.

Also, according to the present invention, it is possible to operate a group of processes that require exclusive control on the same processor,
Normal operation can be guaranteed even if an existing process for a conventional single processor that does not consider exclusive control over shared memory is used without modification. This does not mean that other processors are not used; processes other than the above process groups can be executed on any processor without such restrictions, making effective use of the multiprocessor function. can do.

[Brief explanation of drawings]

Fig. 1 is a schematic diagram of process scheduling according to an embodiment of the present invention, Fig. 2 is a system configuration diagram, Fig. 3 is a diagram of the scheduler algorithm, Fig. 4 is a diagram of another embodiment of the scheduler algorithm, and Fig. 5 Figure 6 is a diagram of an algorithm for assigning a selected process to its own processor; Figure 7 is a diagram of a timer interrupt processing routine; Figure 8 is a diagram of another embodiment of the system configuration. , Figure 9 is an algorithm diagram of the scheduler, Figure 10 is a diagram of a processing routine for timer interrupts every 10 seconds, Figure 11 is a diagram of a processing routine for timer interrupts every second, and Figure 12 is a diagram of the conventional scheduling method. FIG. 13 is an explanatory diagram when a program for a single processor is executed using the scheduling method of the present invention. 4... Connection lot, 5... Shared memory, 11... Processor number, 12... Process, 13... Processor, 14... Timer, 15... Group, 31
...OS program, 32... Process management table, 33... Group management table, 311... Scheduler, 321... Processor identifier, 322...
...Priority, 323...Execution waiting time, 324...
Counter, 325... Processor fixed flag, 326
... Group identifier, 327 ... CPU time, 32
8-330...Priority, 331...Processor identifier, 332...Execution waiting time. 333...Counter, 334...Group fixed flag, 41...Interrupt prohibition processing, 42...Resource use processing, 43...Interrupt enable processing.

Claims (1)

[Scope of Claims] 1. In a multiprocessor system comprising a plurality of processors, the plurality of processors being controlled by a single operating system, a process scheduler of the operating system according to a process management table for managing processes; is a process scheduling method that allocates a processor to a process, in which group identifiers that allow duplication are written in the process management table, and a group management table is created that stores information on processes that have the same group identifier. established,
Processor allocation information that instructs a processor to execute a process in the group is written in the group management table, and the process scheduler refers to the process management table to determine the process for the process selected to be executed next. By determining the group and running the process on the processor specified by the processor allocation information in the group management table of the group, processes belonging to the same group are executed on the same processor. Processor scheduling method. 2. Store the execution waiting time of the process in the process management table, refer to the process management table for the process selected by the process scheduler to execute next, and when the execution waiting time is equal to or greater than a certain value,
When the execution wait time is less than or equal to a certain value when the process is run on an arbitrary processor, the process management table is referred to, the group of the process is determined, and the processor specified by the processor allocation information in the group management table of the group is executed. 2. The multiprocessor scheduling method according to claim 1, further comprising operating said process on a computer. 3. The time during which no process in the group is being executed, that is, the group execution waiting time, is stored in the group management table, and the process scheduler refers to the process management table for the process selected to be executed next. to determine the group of the process,
A multiprocessor scheduling method characterized in that processor allocation information in the group management table can be changed when the group execution waiting time in the group management table for the group is greater than or equal to a certain value. 4. Store the execution waiting time of the process and the identifier of the processor on which the process was last executed in the process management table, and the process scheduler refers to the process management table for the process selected to be executed next. When the waiting time is less than a certain value, the process is executed on the processor that executed it last time or on the processor specified by the processor allocation information in the group management table, and when it is more than a certain value, the process is executed on a processor other than the above. Claim 1 or 2 characterized in that
multiprocessor scheduling method. 5. The multiprocessor scheduling method according to claim 2, 3, or 4, characterized in that the same number of times the process is refused to be allocated to a processor is used instead of the execution waiting time of the process. 6. The multiprocessor according to claim 3 or 5, wherein the number of times assignment of the process to a processor is rejected is used instead of the time during which no process in the group is being executed, that is, the group execution waiting time. Scheduling method. 7. A process management table for managing a process when a processor needs to switch the process being executed in a multiprocessor system including a plurality of processors and the plurality of processors are controlled by a single operating system. According to the process scheduling method, a process scheduler of the operating system is executed to determine the next process to be executed by the processor, a group identifier that allows duplication is written in the process management table, and the group identifier is A group management table for storing information on the same processes is provided, processor allocation information for instructing processors to execute processes in the group is written in the group management table, and the process scheduler selects, from among the executable processes, A multiprocessor scheduling method, characterized in that a process indicated by a group identifier in the process management table and for which processor allocation information in the group management table indicates a processor executing the process scheduler is executed next. 8. Store the execution waiting time of the process in the process management table, and the process scheduler refers to the process management table when selecting the next process to be executed, and when the execution waiting time exceeds a certain value, 8. The multiprocessor scheduling method according to claim 7, wherein the process is made executable. 9. The time during which no process in the group is being executed, that is, the group execution waiting time, is stored in the group management table, and the process scheduler uses the process management table when selecting the next process to be executed. The group of the process is determined by referring to it, and when the group execution wait time in the group management table of the group is equal to or greater than a certain value, the processor allocation information of the group management table is changed to enable the process to be executed. The multiprocessor scheduling method according to claim 7 or 8, characterized in that: 10. Store the execution waiting time of the process and the identifier of the processor on which the process was last executed in the process management table, and the process scheduler refers to the process management table when selecting the process to be executed next. Then, when the waiting time is less than a certain value, only if the processor that executed the process last time or the processor indicated by the processor allocation information in the group management table is the same as the processor that is executing the process scheduler. Claim 7: The process is made executable, and when the waiting time exceeds a certain value, the process is made executable.
multiprocessor scheduling method. 11. The multiprocessor scheduling method according to claim 8, 9 or 10, characterized in that the number of times the process has been refused to be assigned to a processor is used instead of the execution waiting time of the process. 12. The multiprocessor according to claim 3 or 6, characterized in that the number of times the process was refused to be assigned to a processor is used instead of the time during which no process in the group is being executed, that is, the group execution waiting time. Processor scheduling method. 13. In a multiprocessor system that includes a plurality of processors and the plurality of processors are controlled by a single operating system, the process scheduler of the operating system determines which process to use according to a process management table that manages processes. A process scheduling method for allocating processors, wherein the process management table describes priorities for each of the plurality of processors and group identifiers that allow duplication, and stores information on processes having the same group identifier. is provided, processor allocation information indicating a processor to execute a process in the group is provided in the group management table, and processor allocation information in the group management table indicated by the group identifier of the process management table corresponds to the processor specified. A multiprocessor scheduling method characterized by setting high priorities. 14. The multiprocessor according to claim 1, further comprising a code indicating that the process can be executed by any processor as one of the codes of the process allocation information. Scheduling method. 15. Any one of claims 1 to 14, wherein either or both of the process management table and the group management table is provided with a flag indicating that the process can be executed only on a specific processor. The multiprocessor scheduling method described in . 16. Any one of claims 1 to 15, characterized in that exclusive control of the resource is simplified by giving the same group identifier to processes that share the resource.
The multiprocessor scheduling method described in .