Parallel-job scheduling on cluster computing systems
Academic Article
Overview
Additional Document Info
View All
Overview
abstract
Cluster computing system has attracted much attention recently as a new parallel computing model. Because of its heterogeneous, jobs in a cluster computing system may have alternative execution modes and may require parallel execution of multiple resources in the system. This new job execution mode on cluster computing systems has proposed new challenging research projects for system job scheduling and resource management. Based on our previous theoretical research and new investigation on the parallel job scheduling problem, this paper is focused on the study of parallel job scheduling problem on cluster computing systems. We first prove that in a cluster computing system with a large number of processors, it is impossible to develop parallel job scheduling algorithms even with very loose quality requirements. Three new heuristic algorithms LLF (Largest Length First), LWF (Largest Width First) and LAF (Largest Area First) are proposed. The extensive experimental results show that our new algorithms improve significantly over the previous algorithms proposed in the literature. Consequently, these algorithms are more effective in practice for the job scheduling and resource management for cluster computing systems.
