JPH0877029A - Processing request execution order control system based upon load rate - Google Patents

Abstract

PURPOSE: To enable a computer system, consisting of plural computing processors and a server processor to efficiently execute an application program that makes the processors perform processing while synchronizing with one another. CONSTITUTION: A job assigning processor J assigns jobs to the computing processors A-E so that the number of jobs is made uniform. a process on a computing processor sends a processing request message to the server processor S. The server processor S, after receiving the processing request, determines the execution priority level of the processing request in consideration of the load state of the computing processor as the transmission source of the processing request and the scheduling priority level of the process having sent the processing request, executes the processing request according to a determined procedure, and sends a response back to the computing processor. Consequently, the degrees of advance of processings are adjusted among the processors and the program which advances the processings while associating the processors can efficiently be executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an execution order control system for processing requests generated between processors in a computer system in which a plurality of processors cooperate with each other to perform processing.

[0002]

2. Description of the Related Art Conventionally, in a computer system which is composed of a plurality of processors and performs a calculation-centric processing and a server processor which provides a common function to the calculation processors, the processing from the processing of the calculation processor When executing a request on the server processor, the load information of the calculation processor was not reflected in the execution order of the processing request.

An application having a form in which a plurality of processes are spread over a plurality of processors and the processes are synchronized with each other to execute the process is divided into a process (master process) which plays a role of distributing the process to the plurality of processes and a master process. It is composed of processes (worker processes) that undertake processing. When the master process requests processing from a plurality of worker processes, the processing of the master process continues to wait until all worker processes have completed processing and send a response to the master process. For this reason, when worker processes are distributed among multiple calculation processors whose load level varies from processor to processor, the waiting time of the master process is the response time from the worker process on the processor with the highest load, Inefficient latency occurs.

When the master process requests a processing request including I / O processing to all worker processes, all worker processes simultaneously send processing requests to the I / O server processor and wait for a response from the server processor. become. Here, in the conventional method, since the execution order of processing requests in the server processor does not consider the load information of the calculation processor, the response time from the server processor to the calculation processor is uniform regardless of the load of the calculation processor. .

Therefore, in the conventional method, the response waiting time from the worker process of the master process depends on the response time of the worker process on the computational processor with the highest load, and the aforementioned inefficient waiting time is eliminated. There was no effective way to do it.

[0006]

As a method of reducing the difference in load between processors, it is conceivable to distribute the jobs submitted by the user to the respective calculation processors so that the number of jobs becomes even. It is difficult to absorb the dynamic change in the load because the execution time and characteristics of the process may differ for each job, and the number of processes on the calculation processor may change dynamically due to the creation and disappearance of processes. Therefore, when executing a processing request from a computational processor on the server processor, priority is given to the order of execution of processing requests from high-priority processes executing on the computational processor with a high load, We considered to shorten the response time of and to improve the efficiency of the process execution of the process executed on the processor with heavy load.

As a result, when the master process requests a plurality of worker processes for processing requests including I / O processing, and a plurality of worker processes simultaneously requests I / O processing requests to the I / O server processors, the server The response from the processor to the worker process is given priority to the worker process on the computational processor with the highest load. Therefore, the response time from the worker process on the computational processor, which has the highest load, to the master process is shorter than that of the conventional method. Therefore, the response time from the worker process to the master process is made uniform, the time the master process waits for the responses of all the worker processes becomes shorter than in the conventional method, and the inefficient waiting time is improved.

One object of the present invention is to shorten the response time of a server processor for a process executed on a heavy-load computing processor, thereby making the process execution time of the process relatively short. That is.

Another object of the present invention is to efficiently balance an execution time of each process on a calculation processor so that an application program in a form in which the processes are executed in synchronization with each other can be efficiently performed. Is to do.

Still another object of the present invention is to detect the load status of each calculation processor by transmitting the load status together when each calculation processor sends a processing request to the server processor. It is to reduce the overhead for monitoring.

[0011]

According to the present invention, in order to achieve the above-mentioned object, a means for calculating a load of a calculation processor, a server for processing the load, together with a scheduling priority of a given process on the calculation processor are provided. Means for inserting into the processing request message to the processor, and means for calculating the relative load of the calculation processor that has transmitted the processing request from the information included in the processing request message in the server processor, and A means is provided for controlling the execution order of processing requests and the scheduling priority of a server process that executes processing requests based on the obtained load information and the scheduling priority of the process.

[0012]

It is possible to control the execution order of processing requests from the worker process to the server processor and the scheduling priority of the server process that executes the processing requests. The load of the calculation processor and the scheduling priority of the process set by the application can be controlled. The response time from the server processor to the calculation processor can be controlled accordingly. As a result, it is possible to efficiently execute an application program in which processing is performed in synchronization with each other among the processors.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a system configuration diagram of the present invention. A ~
E is a calculation processor.

The calculation processor executes a job submitted by a user as a process. J is a job allocation processor. The job allocation processor has means for grasping the number of jobs allocated to each calculation processor, and allocates jobs so that the number of jobs of each calculation processor becomes equal. S is a server processor. The server processor executes the processing request from the calculation processor. I is an I / O processing device. Here, an I / O server processor is shown as an example of the server processor. N is a high-speed network connecting each processor.

FIG. 2 shows an operation example of an application program operating on the computer system shown in FIG. The jobs submitted to the job allocation processor are shown as jobs J1 and J2. The jobs J1 and J2 are in a state of waiting to be assigned to the calculation processor, and the already assigned job is executed as a process on the calculation processor. The calculation processors B, C, D indicate that other processes B2, C1, D1 not belonging to the master process A1 are being executed. Master process A1 is worker process B
1, a processing request including an I / O processing is sent to E1, and the worker process sends an I / O processing request to the server processor S. In the server processor S, an example is shown in which four server processes S1, S2, S3, S4 are prepared to process the processing request from the calculation processor and the processing request is executed. When there are a number of processing requests that cannot be processed by these four server processes, those processing requests are queued in the processing request queue S5. The processing request that has been processed by the server processor is returned to the worker process of the transmission source as a response. After processing the response from the server processor, the worker process returns the processing result to the master process A1. The processing request from the master process A1 to the worker processes B1 and E1 is shown as A1B1 and A1E1.
1, the processing request from E1 to the server processor S is B1
Denote by S, E1S. The response from the server processor S to the worker processes B1 and E1 is shown as SB1 and SE1, and the response from the worker processes B1 and E1 to the master process A1 is shown as B1A1 and E1A1. In this method, for example, when the load of the calculation processor B is higher than the load of the calculation processor E and the processing requests from the calculation processors B and E are transmitted to the server processor S at almost the same time, the server processor gives priority to the processing request B1S. To execute. This reduces the response time from the worker process on the busiest computational processor to the master process and, as a result, improves the inefficient latency of the master process.

Further, when the load on the server processor is low, or when the number of processing requests is small, the average response time from the server processor to the worker process is short, so the worker when switching the execution order of the processing requests Has little effect on the process. However, when the load on the server processor is high or the number of processing requests is large, the processing request executed first and the processing request executed last are the worker who sent the processing request. Since the response time to the process changes significantly, the effect on the worker process when the execution order of the processing requests is changed is large, and the effect on the master process when the processing requests are executed in an appropriate order using this method. growing.

FIG. 3 is a procedure for creating a processing request message when a process on the calculation processor executes a processing request to the server processor. The load factor of the calculation processor is
It is calculated from the number of runnable processes and the CPU operating rate. The number of runnable processes is the number of processes that share one processor. If the child value is large, the CPU usage time that can be used by one process per unit time will decrease, and the processor load will be high. Represent on the other hand,
The CPU operating rate indicates the rate at which the CPU is operating per unit time. The load factor calculation formula shown in FIG. 3 is shown as an example of the calculation formula for obtaining the load factor in an actual system. Next, a procedure for creating the processing request message will be described. When a process issues a processing request instruction, the system determines the scheduling priority described in the process structure of the process. Next, the CPU operating rate and the number of runnable processes are obtained, and the processor load is calculated using the obtained number of runnable processes and the CPU operating rate. Next, the processing request message shown in FIG. 4 is created, the scheduling priority of the process and the load of the processor are substituted into the message, and the message is sent to the server processor.

FIG. 5 shows a procedure for determining the execution order of processing requests in the server processor. The processing request priority (execution order) is obtained by the server processor in order to determine which processing request is executed first. The priority of the processing request is obtained from the scheduling priority of the worker process that has transmitted the processing request and the relative load factor of the computing processor executing the process. The relative load factor of the calculation processor is calculated by (load factor of the calculation processor / average of load factors of all calculation processors that have transmitted the processing request). The calculation formula of the processing request priority shown in FIG. 5 is shown as an example of the calculation formula for obtaining the priority of the processing request in the actual system. Next, a procedure for determining the execution order of processing requests will be described. First, the processing request message is received, and the source address of the processing request, the scheduling priority of the process that sent the processing request, and the load factor of the calculation processor are recorded in the execution order management table shown in FIG. Next, referring to the process scheduling priority and the processor load in the execution order management table, the processing request priority is obtained from the calculation formula of the processing request priority shown in FIG. 5, and recorded in the execution order management table. The processing request transmission source address shown in FIG. 6 indicates an address by which the processing request can be centrally identified on the entire system, and the process scheduling priority is the scheduling priority of the worker process that has transmitted the processing request. Used to determine the scheduling priority of the server process that executes the processing request above. Further, the processor load indicates the load factor of the calculation processor that has transmitted the processing request, and the processing request priority indicates the priority obtained from the calculation formula of the processing request priority shown in FIG. Next, based on the processing request priority of the execution order management table, the processing request is inserted into the relevant part of the processing request queue for temporarily holding the processing request. The processing request queue is a multistage queue as shown in FIG. Each stage of FIG. 7 shows the processing request priority, and corresponds to the processing request priority in the execution order management table of FIG. The execution order is such that the processing requests queued in the higher priority stage are executed in order. As used in conventional UNIX process scheduling, a processing request of a lower priority is moved to a higher level after a certain time has passed, and it is always executed after a certain time has passed. Be guaranteed.

FIG. 8 shows an execution procedure of the processing request.
Of the four server processes on the server processor, the server process in the standby state takes out the processing request queued at the highest level in the processing request queue, and the priority of the server process is given to the process scheduling priority in the execution order management table of FIG. Change processing every time and execute processing request. After the execution is completed, a response message composed of the destination address, the source address, and the processing result is created and transmitted to the source calculation processor.

[0020]

As described above, according to the present invention, the response time of the server processor can be controlled according to the state of the transmission source calculation processor, so that the processing request is executed while ignoring the state of the calculation processor as in the conventional method. Compared with the case, when the load varies between the calculation processors or when the load on the server processor increases, it is possible to efficiently execute the application program in the form of executing a plurality of processors in cooperation with each other.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of the present invention.

FIG. 2 is an operation example of an application program operating on the system of the present invention.

FIG. 3 is a diagram showing a procedure for creating a processing request message.

FIG. 4 is a configuration diagram of a processing request message.

FIG. 5 is a diagram showing a procedure for determining an execution order of processing requests.

FIG. 6 is a configuration diagram of an execution order management table.

FIG. 7 is a diagram showing a processing request queue.

FIG. 8 is a diagram showing an execution procedure of a processing request.

[Explanation of symbols]

A, B, C, D, E ... Computational processor, J ... Job allocation processor, S ... Server processor, I ... Disk device, N ... High-speed network, A1 ... Master process, B1, E1 ... Worker process, B2, C1, D1 ... Process not belonging to master process A1, J1, J2 ... User job, S5 ... Processing request queue, S1, S2, S3, S4 ... Server process, A1B1, A1E1 ... Processing request from master process to worker process, B1A1, E1A1 ... Response from worker process to master process, B1S, E1S ... Processing request from worker process to server processor, SB1, SE1 ... Response from server processor to worker processor, 71, 72, 73 ... High priority processing request, 74, 75 ... Low priority processing request.

Claims (1)

[Claims] 1. A computer system in which a plurality of processors each comprising a memory and a CPU are coupled together, and a computer processor which mainly performs CPU-dependent calculation processing,
A server processor that provides common functions to each processor such as I / O processing and network processing, and that executes processing requests from the calculation processor, and a job allocation processor that evenly allocates jobs submitted by users to the calculation processors. The calculation processor has means for detecting the load status of its own processor, and when a process on the calculation processor notifies the server processor of a processing request, the scheduling priority of that process and the processor When the server processor executes a processing request, it has a means to notify the load status, and controls the execution order and execution priority based on the scheduling priority of the process that sent the processing request and the load status of the processor. A process request execution order control method characterized by the following.