Optimal Size of Job Pool for Initiating a Scheduling Event
Abstract
In today's military with its dwindling resources, making the best use of computers, particularly to support real time commercial off the shelf (COTS) applications, is becoming critical for success. Resource Management Systems (RMS) strive to address this issue. The RMS's job scheduler is needed to ensure good quality of service (QoS) to all applications. This research uses discrete event simulation experiments to investigate the cost tradeoff between improving system performance through grouping incoming jobs to create better schedules, versus both: (1) the time spent waiting for the group to accumulate, and (2) the additional cost of computing schedules involving more jobs. A MaxMin O(MN(2)) greedy scheduling algorithm attempting to minimize the total time in system was used in these experiments. We analyzed the data generated from numerous experiments that used typical input parameters. As a result of this effort, we conclude that job grouping should be used when the utilization factor for the system is near 1.0, or precisely when the mean arrival rate is comparable to the total mean service rate of the processors. At this utilization rate, the group size should be equal to the number of machines in the system. However, when the utilization factor is significantly different from 1.0, each job should be scheduled as it arrives.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1999
- Accession Number
- ADA370839
Entities
People
- James M. Breitinger
Organizations
- Naval Postgraduate School