Job Management in Grids of MOSIX Clusters

摘要

EnFuzion and MOSIX are two packages that represent different approaches to cluster management. EnFuzion is a user-level queuing system that can dispatch a predetermined number of processes to a cluster. It is a commercial version of Nimrod, a tool th at supports parameter sweep applications on a variety of platforms. MOSIX, on the other hand, is operating system (kernel) level software that supports preemptive process migration for near optimal, cluster-wide resource management, virtually makings the cluster run like an SMP. Traditionally, users either use EnFuzion with a conventional cluster operating system, or MOSIX without a queue manager. This paper presents a Grid management system that combines EnFuzion with MOSIX for efficient management of processes in multiple clusters. We present a range of experiments that demonstrate the advantages of such a combination, including a real world case study that distributed a computational model of a solar system.