Job Management System For Cluster

Computer development has undergone five revolutions from 1945, when the modern computer era, until now Massive Parallel Processing is the most important character of the fifth generation of computers. Because of its outstanding price/performance radio and scalability, cluster system is becoming the trend of Computer Architecture and Parallel Processing. It is completely able to satisfy our requirement that keeps increasing to computer performance unceasingly. Clusters of workstations are increasingly viewed as a cost-efficient alternative to parallel supercomputer. It is the hotpot of current research.Job Management System (JMS) is critical to give full play to the performance of cluster and improve the rate of its utilization. It can make reasonable use of the system resources and expensive CPU time, effectively manage cluster system, and equitably schedule jobs. Thus, many researchers show solicitude for Job Management for Cluster.This thesis investigates the essential issues of job management for cluster, proposes one novel algorithm for job scheduling to increase the throughput of a cluster, and introduces the job management system JOSS we developed for Dawning 2000, which is cluster-based computer. It includes the following parts:1. The basic concepts, components, functions and characters of job management system for cluster are presented.2. Five representatives of JMS for MPPs and clusters, such as Loadleveler, LSF, are introduced. Their basic structures and characteristics are analyzed. In addition, the evaluation criterion of JMS is presented.