JPH03148730A - Process queue processing system - Google Patents

Abstract

PURPOSE:To decrease access to a shared memory and to control a process by a process queue to be efficiently generated by linking the queue of a required process group by a control block sub link. CONSTITUTION:A process queue 14 connects a queue 15 of the required process group to an individual control block 16 and successively links the control blocks 16. When the process queue is added, a link relation between queue elements to be added is simultaneously set and connected to the control block 16 in the end of the existent queue. Namely, for a control program to be executed by a required arithmetic unit 11, the added part of the process queue 15 is prepared on a main storage 12 and afterwards transferred to a shared memory 13 and when only one link of the control block 16 in the end is updated, the queue can be generated on the shared memory 13 by twice access at a maximum. Thus, the access to the shared memory is decreased and the process can be controlled by the process queue to be efficiently generated.

Description

[Detailed Description of the Invention] [Summary] Regarding the queuing of processes executed in a multiprocessor system consisting of an arithmetic unit having a shared storage device and a separate main storage device, efficiency is improved by reducing accesses to the shared storage device. The objective is a process queue processing method in which processes are controlled by process queues that can be easily generated, and includes a plurality of arithmetic units, a main storage device, and a shared storage device, and each of the arithmetic units includes:
The individual main storage devices are connected to the common shared storage device, and each processing unit takes out a process of a designated process group from the process queue placed on the shared storage device and processes the main storage device. The process program loaded in the storage device is executed, and the process queue is formed by connecting the queues of the required process groups to individual control blocks and sequentially linking the control blocks. The queue is configured to be a queue of the process for each process group.

[Industrial Icheon field]

The present invention relates to a process queue processing method executed in a multiprocessor system consisting of an arithmetic unit having a shared storage device 4 and an individual main storage device.

[Prior Art] FIG. 2 is a block diagram showing an example of the configuration of a multiprocessor system.

The plurality of processing devices 1 are connected to individual main storage devices 2, and each processing device 1 executes programs separately loaded into the main storage device 2 and stored in the main storage device 2 or the shared storage device 3. access data to perform necessary system management and applied business processing.

The shared storage device 3 can be accessed by all the computing devices 2, and contains areas for commonly used management data and work data for business processing.

With this system configuration, for example, a program is loaded into the main memory device 2 of a plurality of necessary arithmetic devices, and a task executed by each arithmetic device 1 can partially interrupt the execution of the program according to management information provided in the shared storage device 3. By this,
A single program can be divided into appropriate configurations and executed in parallel by multiple tasks.

For this purpose, when a compiler generates a target program, it divides the program into processes, and further divides the program into processes that should be executed on the same arithmetic unit W1 to form a process group, and divides these processes into processes for each of the arithmetic units. Make tasks run sequentially.

Note that each arithmetic unit l may have multiple tasks,
Further, each arithmetic device may be composed of a plurality of processing units, in which case each arithmetic device can also execute a plurality of tasks in parallel.

In order to control the above-mentioned processes, for example, at the start of a so-called job step, a management program executed on a certain computing device 1 sends a queue 9 to which process control blocks (PCBs) of processes of programs to be executed in that job step are linked. is generated on the shared storage device 3, and for example, when the task of the arithmetic unit l starts executing a process according to the PCB, and the program reaches the execution of a predetermined parallel processing call instruction, the parallel processing is executed. Typically multiple queues 9 for the required processes are added on the shared storage 3 and processes are added to the required queues 9.

In each processing device 1, according to the process group number specified separately, the task of each processing device Wi takes out a PCB connected as a queue element from the queue 9 of the corresponding process group, and executes the program of the process according to the information. do. Each PCB has a configuration, for example, as shown in FIG. Consists of first name, etc.

[Problem to be solved by the invention]

The shared storage device 3 requires a relatively large capacity N (for example, several gigabytes) to be able to store a large amount of data, is configured with relatively low-speed storage elements due to economical requirements, and is not capable of transferring blocks of large amounts of data. It is configured to process efficiently, and the generation of the queue 9 as described above and the PC input to the queue are
There is a problem in that access for adding B and the like is relatively inefficient, and that creating and adding queues for each process group requires a relatively long processing time.

An object of the present invention is to provide a process queue processing method that controls the execution of processes using a process queue that can be efficiently generated by reducing access to a shared storage device.

[Means to solve the problem]

FIG. 1 is a block diagram showing the configuration of the present invention.

The figure shows the configuration of a process queue processing system, which includes a plurality of arithmetic units 11, a main storage device 12, and a J-recording device 1.
3, and each arithmetic unit 11 has an individual main memory 12
and a common shared storage device 13, each computing device 1
1 is a block queue 1 placed on the co-located storage device 13.
4, the process of the specified process group is taken out and the program of the process loaded into the main memory 12 is executed, and the process queue 14 connects the queue 15 of the required process group to the individual control block 16. The control blocks 16 are sequentially linked, and the queue 15 of each process group is a queue of the process for each process group.

[For production]

With the above processing method, for example, at the start of parallel processing,
When adding a process queue containing the processes necessary for parallel processing, if the queue configuration of the present invention is used, all the link relationships between the added queue elements can be set at once, and the end of the existing queue can be controlled. Since the management program executed on the required processing unit 11 creates an additional part of the process queue 15 on the main storage device t2, it transfers it to the shared storage device 13. However, by updating only one link in the control block 16 at the end, a queue can be efficiently generated on the shared storage device with at most two accesses. When accessing a queue, the required process group is controlled by the control block 16.
Since the search can be performed by scanning the queue 15 of the required process group, the efficiency of accessing the queue 15 of the required process group does not deteriorate significantly, and the processing efficiency of the system can be improved overall.

〔Example〕

In the system shown in FIG. 1, when a program executed by a certain arithmetic unit 1 reaches the execution of a parallel processing call instruction,
Now, in order to add queues for the plurality of processes necessary for the parallel processing to the shared storage device 3, a prototype of an additional part of the process queue 14 is created on the main storage device 12.

That is, first, a process group queue 15 is created in the main memory x12, in which the processes of each process group constituting the parallel processing part are linked. The queue 15 of each process group is
A PCB shown in FIG. 3(a) similar to the conventional case is configured as a cue element.

The queues 15 of the process group are then each connected to a separate control block 16. The control block 16 is, for example, a third
With the configuration shown in Figure (B), the head of the process group queue 15 is indicated by a PCB link, and the control blocks 16 are linked and connected by sublinks to the process group queue 1.
It consists of the same process group number column as 5 and other information.

In this way, by linking the queues 15 of the necessary process groups with the sublinks of the control block 16 and creating a prototype of the additional part of the process queue 14 in a contiguous area of the main storage 12, it can be shared all at once. Memory device W13
to the specified area.

Here, the sublink of the last control block 16 has a predetermined symbol indicating the end, so next the shared storage device f1
Adding a queue to the process queue 14 on the shared storage device 13 by updating the value of the sublink of the tail control block 16 of the previous queue on 3 to point to the head control block 16 of the added queue. is completed.

In each processing unit 1, in accordance with a separately specified process group number, the task of each processing unit 21 first traces the control block 16 of the process queue 14 according to its sublinks in order to search the queue 15 of the corresponding process group. ,
The required queue is determined by the process group number held in each control block 16, and the queue I of that process group is determined.
5, and execute the program of the process according to the information.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, in a multiprocessor system consisting of an arithmetic unit having a shared storage device and a separate main storage device, accesses to the shared storage device can be reduced and generation can be performed efficiently. Since the process is controlled by the process queue, it has a significant industrial effect of improving the processing efficiency of the system.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of the present invention, FIG. 2 is a block diagram showing an example of a conventional configuration, and FIG. 3 is an explanation of one cue element. In the figure, 1.11 is an arithmetic unit, 2.12 is a main storage device, and 3
．． 13 is a shared storage device, 9 is a queue, 14 is a process queue, 15 is a process group queue, and 16 is a control block. Patent applicant: Director of the Agency of Industrial Science and Technology Ken Sugiura J
1,3 LIe116 row 1 control block Block diagram showing the configuration of the present invention Figure 1

Claims (1)

[Claims] It has a plurality of arithmetic units (11), a main memory device (12), and a shared memory device (13), and each of the arithmetic units (11) has a separate main memory device (12). )
and the common shared storage device (13), and each computing device (11) selects a process of a designated process group from the process queue (14) placed on the shared storage device (13). and executes the program of the process loaded in the main memory (12), and the process queue (14) stores the queue (15) of the required process group as an individual control block (16). 1. A process queue processing method, characterized in that the control blocks are sequentially linked to each other, and the queue (15) of each process group is a queue of the process of each process group.