JPH0612394A - Process schedule system - Google Patents

Abstract

PURPOSE:To perform exclusive control over a resource without any overhead of process switching by limiting the number of processes to be spun at the same time, prevent processor capability from wastefully being used owing to the spinning, and entering a resource wait state by a spinning system when the number of competition is small. CONSTITUTION:When requests to use a resource in a resource control table 3 are made from processes represented in process control tables 2a-2d at the same time, one process is allowed to use the resource and a predetermined number of processes from among the remaining processes are placed in the resource wait state by the spinning system; when there are still processes which are not included, the processes are suspended and switched to processes which do not other resources, thereby carrying on the processing on a processor.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process scheduling method in a multiprocessor system sharing a memory.

[0002]

2. Description of the Related Art As a process scheduling method in the case of resource competition in a multiprocessor system sharing a memory, conventionally only one process uses the resource, and the remaining processes loop on the processor to wait for the release of the resource. A method (spin method) and a method (suspend method) of interrupting the processing of the remaining processes and switching to another process not requesting the resource and executing the method have been adopted.

Further, as disclosed in Japanese Patent Laid-Open No. 1-2671771, when a spin method is used for waiting for a normal resource and a process using the resource interrupts the processing due to an interrupt process or the like, as an improvement plan of the spin method. Only, a method of interrupting the process requesting the resource is also proposed.

[0004]

Among the conventional methods, in the spin method, the processor does not perform any effective processing during the spin, and if the resource is occupied for a long time, the waste of the processing capacity of the processor is neglected. Can not. Further, in the suspend method, when the overhead required for process switching is large compared to the time for occupying resources and performing processing, waste of processing capacity cannot be ignored.

It is an object of the present invention to provide a process scheduling method that reduces the overhead required for process switching.

[0006]

To solve this problem, as an extension of the spin method, a method has been proposed in which the number of processes waiting for spin is limited, and a certain number or more of processes are suspended by the suspend method. .

[0007]

Function: The waiting overhead of the spin method is limited. The maximum value of the number of processes to be spun is set so that the overhead of continuously spinning is larger than the overhead of switching processes by suspending.

[0008]

EXAMPLE An example of the present invention will be described below. FIG. 1 shows the outline of the present invention. The process is managed by the processor management tables 2a to 2d on the main memory. The process scheduler 5 associates any of the process management tables 2a to 2d with the processors 1a to 1b,
Causes the processor to execute the process. The resource management table 3 is a table for managing resources and includes a resource status 8 (set to “in use” or “unused”). The process states 6a to 6d in the process management tables 2a to 2d are respectively "resource unused execution", "resource usage executing", "resource unused execution waiting", "resource usage execution waiting spin", and "resource usage". It is set to one of the five states of "suspend for execution". Further, a scheduler lock 7 (set to “in use” or “not used”) is used to allow only one processor to process the process scheduler 5 at the same time.

FIG. 2 shows a process flow of the process scheduler when a process requests a resource. First,
A scheduler lock reservation subroutine, which will be described later, is called to reserve a scheduler lock (block 211).
Next, it is checked whether the resource is being used by another process (block 212), and if it is not being used, the resource state is changed to being used (block 211), and the state of the own process is changed to "resource in use executing". (Block 222), sets the scheduler lock to unused (block 223), and continues execution of the process (block 224). If the resource is in use, it is checked whether the number of processes waiting in the "resource use execution waiting spin" state is smaller than the predetermined process spin upper limit number (block 213).
If the former is small, the process state is set to the "resource use execution waiting spin" state (block 214), the scheduler lock is set to unused (block 215), and after spinning for a predetermined time (block 216), the block Returning to step 211, after securing the scheduler lock again, it is checked whether the resource is released. In block 213, if the former is greater than or equal to, the process is interrupted, and if it is restarted, it is set to restart the process from the start point of this routine (block 201), and the process state is set to "resource use execution". It is set to "suspend wait" (block 202), and it is checked whether there is any other process whose process status is "waiting for resource unused execution" (block 203). If there is no such process, the scheduler lock is set to unused (block 217) and the processor goes idle (block 218). If there is a process that meets the conditions of block 203, the status of the found process is changed to "resource in use" (block 204), the scheduler lock is set to unused (block 205), and the process is executed. Begin (block 206).

FIG. 3 shows the flow of processing when a process using a resource releases the resource. First, a scheduler lock is secured by calling a subroutine (block 311). Next, it is checked whether there is another process whose status is "spin for resource use execution" (block 312). If there is such a process, the status of the own process is changed to “resource unused execution”, and the process is executed next (block 31).
5). Furthermore, if there is an idle processor (block 314) and there is a process whose status is "waiting for resource unused execution" (block 315), the status of the found process is changed to "resource unused running". ,
Cause one of the idle processors to begin execution (block 316). Then, the process returns to block 314 and the check is repeated. If there is no processor or process that meets the conditions in block 314 or block 315, set the scheduler lock to unused (block 3
01), the processing of the process which is first decided to be executed next is started (block 302). Also, block 31
If there is no process satisfying the condition in 2, then it is checked whether or not there is a process whose process status is "suspension of use of resource waiting" (block 321). If there is no process, the process proceeds to block 313. If there is such a process,
The state of the found process is changed to "resource in use" and the next process is executed (block 322).
The state of the original process is changed to "waiting for resource unused execution" (block 323), and the process goes to the check in block 314.

FIG. 4 shows a processing flow of a subroutine for securing a scheduler lock. First, the status of the scheduler lock is checked, and if it is not used, it is changed to being used (block 401). This check and setting process is performed by Tes which is generally used in a multiprocessor sharing a memory.
It must be executed without being interrupted by another processor, such as by a t and Set instruction. If the scheduler lock is unused and the above process is successful, the process directly returns to the call source (block 40).
3). If the scheduler lock is in use and the above process fails, the setting of block 401 is tried again after spinning for a predetermined time (block 402).

[0012]

Since the number of processes that can enter the spin wait for the resource at the same time is limited, the spin wait for a long time is prevented, and the database processing can be executed with an appropriate overhead.

[Brief description of drawings]

FIG. 1 shows a system configuration diagram of a process scheduler.

FIG. 2 shows a process flow when a process resource is requested.

FIG. 3 shows a process flow when releasing resources of a process.

FIG. 4 shows a flow of processing when securing a scheduler lock.

[Explanation of symbols]

1a-1b: Processor, 2a-2d: Process management block, 3: Lock management table, 4: Main memory, 5:
Process scheduler.

Claims (1)

[Claims] 1. A multiprocessor that shares a main memory, has a plurality of processes and a process scheduler that schedules the processes, and each process alternates a process that uses only one resource and a process that does not use it. When multiple processes running on the processor try to use the resource at the same time, only one process uses the resource for processing, and among the remaining processes, the predetermined process is performed. A process with a maximum number of processes waits for resources to be used on the processor in a loop, and the remaining processes that exceed the maximum number of processes stop processing and search for a process that does not use another resource, and switch to this process if any. A process scheduling method characterized by continuing processing on a processor.