JPS62184543A - Job scheduling system - Google Patents

Abstract

PURPOSE:To prevent a throughput from becoming worse by sending the job of a class, until the multiplicity of which reaches to the multiplicity set by every job class queue classified by scales to a job executing part, and setting the multiplicity at a larger number for a smaller scale of job. CONSTITUTION:The scale of a thrown-in job 11 is decided from the scheduled time to use a CPU in an executing job administration language at a scale deciding part 12, and the job is registered to either of a large class of queue 13, a medium class of queue 14, or a small class of queue 15 corresponding to the scale of the job. Each of a large class of execution deciding part 16, a medium class of execution deciding part 17, and a small class of execution deciding part 18 sends the job until the multiplicity of the job reaches to the one set for the class to a job executing part 19. In such a case, the multiplicity set for the large class of job is smaller, and for the class medium, larger than that set for the large class of job, and for the small, larger than the medium class. The job in the job executing part 19 is executed under a multiprogramming circumstance.

Description

[Detailed Description of the Invention] "Industrial Application Field" This invention is one of the functions of the operating system of a computer.
This relates to a job scheduling method for selecting jobs to be executed.

"Conventional technology" - In the conventional job scheduling method, a number of job classes with execution multiplicity set in advance are prepared, and the job submitter selects one of these job classes using parameters in the execution JCL (job management language). I was supposed to.

[Problems to be Solved by the Invention] In the conventional job scheduling method described above, jobs of various sizes are scattered in several classes to which a predetermined multiplicity is assigned, and these jobs are Regardless of how many programs of any size are being executed in the system at that time, they will be executed to the full multiplicity of the class to which they belong, resulting in the following drawbacks.

In other words, if jobs with long CPU (Central Processing Unit) execution times are executed at the same time, jobs waiting to be executed may be forced to wait until one of the preceding long-running jobs is finished, even if the CPU execution time is short. This results in the problem of not being able to maintain a proportional relationship between the size of the job and the turnaround time (the time from issuing a request for job processing to processing and returning the results), and also because memory or temporary files are If only jobs that use a large amount are executed at the same time, competition for resources such as memory or temporary files (contention between the two jobs occurs).

As a result, a problem arises in that the throughput (throughput within a certain period of time) of the system is reduced.

``Means for Solving the Problems'' The job scheduling method of the present invention has a job class queue based on the job size (two job class queues), and the execution multiplicity of the job class queues is set to Set small classes to be large,
The execution JCL of the submitted job is analyzed to determine the scale of the job, and the submitted job is registered in a job class queue corresponding to the scale.

Embodiment Next, the present invention will be described with reference to the drawings. FIG. 1 is a schematic diagram of an embodiment of the invention. The size of the job is determined from the CPU time scheduled to be used in the job management language), and the job is registered in one of the five large class queues 13, medium class queues 14, or small class queues 15 according to the size of the job ('''). be done.

In the scale determination section 12, for example, as shown in FIG.

Check whether the job has been submitted (S,), and when the job has been submitted, check the scheduled CPU time used (S2)
), if the time X is 1 hour or more, register the job in the large class queue.
If the time X is within 10 minutes, the job is registered in the middle class queue (S, t), and if the time X is less than 10 minutes, the job is registered in the small class queue (S5).

The large class execution determination section 16, the middle class execution determination section 17, and the small class execution determination section 18 each send jobs to the job execution section 19 until the multiplicity set for the class is reached. In this case, the set multiplicity is small for large-class jobs, thicker than the large-class job setting for medium-class jobs, and further (2) larger for small-class jobs.

Jobs in the job execution unit 19 are executed under a multiprogramming environment.

The execution determination units 16, 17, and 18 perform processing as shown in FIG. 3, for example. In other words, the large class job has reached the set multiplicity (it is checked whether the two-person class job is running (Ss), and if the set multiplicity has not reached the full set multiplicity, there are jobs waiting to be executed in the large class queue 13. It is checked whether there is a job waiting to be executed (S7), and if there is a job waiting to be executed, one of them is executed (S8).Similarly, it is checked whether the middle class job is running at the set multiplicity (ζ2).
S9), and if it is not full, it is checked whether there are any jobs waiting to be executed in the middle class queue 14 (S9).
lo).

If there are jobs waiting to be executed, execute one of them (Sl,
). Also, it is checked whether the small class jobs are running at the set multiplicity (S12), and if it is not full, it is checked whether there are any jobs waiting to be executed in the small class queue 15 (S13). , if there is a job waiting for execution, one of them is executed (S14).

"Effects of the Invention" As explained above, the job scheduling method according to the present invention has job class queues for each job size.
The execution multiplicity of these classes is set to be as small as possible for classes with large jobs and as large as classes of jobs that are small, and the scale judgment unit analyzes the execution JCL of the submitted job and determines the size of the job. By determining the job size and registering the job in the corresponding class queue (= registering the job, the smaller the job size, the shorter the turnaround time, and conversely, the larger the job size, the longer the job turnaround time.) This has the effect of maintaining the proportional relationship between jobs, and prevents jobs from competing for resources due to large-scale jobs being executed at the same time, which would cause the jobs to compete with each other and reduce system throughput.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an outline of an embodiment of the present invention, and FIG. 2 is a diagram showing a flow of processing by a scale determination section in the embodiment. FIG. 3 is a diagram showing the flow of processing of the execution determination section in the embodiment.

Claims (1)

[Claims] (1) A size determining means for determining the size of an submitted job; a plurality of size-based job class queues in which jobs are registered according to the size of the job determined by the size determining means; means for sending jobs of that class to the job execution unit until a set multiplicity is reached for each job class queue, and a job schedule in which the multiplicity is set to be larger as the size of the job is smaller method.