Hierarchical scheduling method for processing tasks having precedence constraints on a parallel processing system

• US 5,392,430 A
• Filed: 10/30/1992
• Issued: 02/21/1995
• Est. Priority Date: 10/30/1992
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines SA (International Business Machines Corporation)
• Inventors: Wolf, Joel L., Yu, Philip S., Chen, Ming-Syan, Turek, John J. E.
• Primary Examiner(s): Heckler, Thomas M.
• Application Number:
  US07/968,717
• Time in Patent Office: 844 Days
• Field of Search: 395/600, 395/650, 364/DIG. 1, 364/DIG. 2 MS File
• US Class Current: 718/102
• CPC Class Codes: G06F 8/451 Code distribution considering CPU load at run-time G06F9/505; load rebalancing G06F9/5083
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching

Abstract

A plurality of queries (jobs) which consist of sets of tasks with precedence constraints between them are optimally scheduled in two stages of scheduling for processing on a parallel processing system. In a first stage of scheduling, multiple optimum schedules are created for each job, one optimum schedule for each possible number of processors which might be used to execute each job, and an estimated job execution time is determined for each of the optimum schedules created for each job, thereby producing a set of estimated job execution times for each job which are a function of the number of processors used for the job execution. Precedence constraints between tasks in each job are respected in creating all of these optimum schedules. Any known optimum scheduling method for parallel processing tasks that have precedence constraints among tasks may be used but a novel preferred method is also disclosed. The second stage of scheduling utilizes the estimated job execution times determined in the first stage of scheduling to create an overall optimum schedule for the jobs. The second stage of scheduling does not involve precedence constraints because the precedence constraints are between tasks within the same job and not between tasks in separate jobs, so jobs may be scheduled without observing any precedence constraints. Any known optimum scheduling method for the parallel processing of jobs that have no precedence constraints may be used, but a novel preferred method is also disclosed.

Claims

1. A method of scheduling a multiplicity of tasks having precedence, constraints on a plurality of processors operating in parallel, comprising the steps of:

(a) defining a plurality of jobs, each of said jobs comprising a portion of said tasks and precedence constraints relating pairs of said tasks which are included only within a single job;

(b) for each said job, creating a plurality, of task schedules for said tasks of said job, each of said task schedules corresponding to a different number of said processors which might possibly be allotted to said job and respecting any precedence constraints among said tasks of said job;

(c) determining an estimated job execution time for each of said task schedules;

(d) using said estimated job execution times for each of said jobs and for each different number of processors which might be allocated to each of said jobs, determining an allotment of processors for each of said jobs;

(e) creating a job schedule for said jobs using said determined allotments; and

(f) executing said jobs on said processors using a job schedule created in step (e).

2. A method as defined in claim 1 wherein said jobs defined in step (a) are each defined as a database query.

3. A method as defined in claim 1 wherein step (b) includes the step of ordering said tasks within said job in a manner which respects precedence constraints for said tasks of said job.

4. A method as defined in claim 3 wherein step (b) further includes the step of determining for each possible number of processors from one to all of said plurality of processors and for each said task within said job a minimum completion time for said task while respecting precedence constraints among said tasks.

5. A method as defined in claim 1 wherein step (d) includes the step of computing the product of estimated job execution time and number of processors corresponding thereto for each of said task schedules and an initial allotment of processors for each of said jobs corresponds to a minimum one of such computed products for each of said jobs.

6. A method as defined in claim 1 wherein said job schedule created in step (e) is created using a two dimensional bin packing method.

7. A method as defined in claim 1 and further comprising the step of analyzing said job schedule created in step (e) for idle processor time within any of said jobs in said job schedule, and if there is such idle processor time, refining said allotments determined in step (d) and creating a new job schedule based on said refined allotments.

8. A method as defined in claim 7 wherein said refining of allotments includes an increase by one processor in said allotment of processors to said job having a maximum amount of idle processor time within said job.

9. A method for scheduling a plurality of separate jobs on a computing system having a multiplicity of processors operating in parallel, each of said jobs comprising at least one task and at least some of said jobs consisting of a plurality of tasks among which there may be precedence constraints, said processors being each capable of executing a separate task simultaneously, comprising the steps of:

(a) creating a set of estimated job execution times for each job, each said estimated job execution time in any said set corresponding to a different number of said processors which might possibly be dedicated to the execution of said each job, said estimated job execution times being created while obeying any precedence constraints for tasks which comprise said each job;

(b) for each said estimated job execution time, computing the product of job execution time and number of processors corresponding thereto;

(c) for each said set of estimated job execution times, identifying a minimum one of said computed products therefor and tentatively assigning said number of processors corresponding to said identified minimum product for execution of said each job;

(d) creating a tentative overall schedule for execution of all of said jobs in parallel by applying a two dimensional packing algorithm to said tasks within each of said jobs using said tentatively assigned number of processors for each said job;

(c) estimating an overall execution time corresponding to said tentative overall schedule and recording said estimated overall execution time;

(f) determining whether there is any idle processor time in said tentative overall schedule;

(g) if so, identifying a job in said overall schedule which has less than a maximum number of processors tentatively assigned for execution of said identified job and which is associated with a maximum amount of idle processor time;

(h) increasing the number of processors tentatively assigned for execution of said identified job;

(i) repeating steps (d)through (f) until said idle processor time in said tentative overall schedule cannot be further reduced thereby; and

then

(j) selecting a minimum overall execution time from said recorded overall execution times and using said tentative overall schedule which corresponds to said minimum recorded overall execution time for executing said jobs in parallel on said processors.