A SIMULATION BASED PERFORMANCE ANALYSIS OF SCHEDULING IN A PARALLEL SYSTEM

摘要

This work addresses performance issues associated with job scheduling in a shared memory partitionable parallel system. Four processor scheduling policies and one I/O scheduling strategy are used to schedule parallel jobs for a variety of workloads. Two cases of I/O architecture are considered. In one case there is a single I/O node, while at the other case there are two independent parallel I/O nodes. Based on a simulation model the performance of the different scheduling policies is investigated. Simulation results reveal that the scheduling policy that schedules the job with the largest number of parallel tasks first performs better than the other policies examined.