Minimizing Paging Tradeoffs Applying Coscheduling Techniques in a Linux Cluster

Abstract

Our research is focused on keeping both local and parallel jobs together in a non-dedicated cluster or NOW (Network of Workstations) and efficiently scheduling them by means of coscheduling mechanisms. The performance of a good coscheduling policy can decrease drastically if memory requirements are not kept in mind. The overflow of the physical memory into the virtual memory usually provokes a severe performance penalty. A real implementation of a coscheduling technique for reducing the number of page faults across a non-dedicated Linux cluster is presented in this article. Our technique is based on knowledge of events obtained during execution, such as communication activity, page faults and memory size of every task. Its performance is analyzed and compared with other coscheduling algorithms.

This work was supported by the MCyT under contract TIC 2001-2592 and partially supported by the Generalitat de Catalunya -Grup de Recerca Consolidat 2001SGR- 00218.