Efficient job scheduling in a mesh multicomputer without discrimination against large jobs

  • Authors:
  • Dugki Min;M. W. Mutka

  • Affiliations:
  • -;-

  • Venue:
  • SPDP '95 Proceedings of the 7th IEEE Symposium on Parallel and Distributeed Processing
  • Year:
  • 1995

Quantified Score

Hi-index 0.00

Visualization

Abstract

Many innovative schemes for allocating jobs to parallel computing systems have been proposed in order to achieve highly utilized parallel computing systems. The schemes have tried to achieve good job response times with little system fragmentation of processing resources. Since most schemes have concentrated on approaches for processor allocation, the schemes have used First-Come-First-Serve (FCFS) as the job scheduling discipline. However, it has been previously established that job scheduling algorithms for parallel computing systems can have a large impact on the system utilization and job response time. Schemes that use multiple queues, which reorder the sequence of jobs allocated to the parallel system, can be very effective in improving the system performance. However, such non-FCFS schemes have been criticized because they provide improved average performance by favoring small jobs at the expense of large jobs. In order to achieve improved performance by means of multiple queue job scheduling schemes without sacrificing the fairness of FCFS, we propose a new job scheduling discipline that behaves in a FCFS manner under low loaded conditions, but exploits performance enhancing features of multiple queue schemes under highly loaded conditions. In addition, the scheme does not inappropriately discriminate against large jobs.