Processmigreringssystem Ett effektivt sätt att utnyttja ett beräkningskluster?

Abstract

The construction and use of clusters to replace ?supercomputers? are becoming more and more common. The cluster systems that are constructed are often based upon a message passing system or a batch system that statically place the processes on different computers. We believe that the clusters of the future will be based upon some kind of process migrating system. Clusters will then work and look like a single system and behave in the same way as a traditional ?supercomputer?. The purpose of this thesis is to compare two process migrating systems for the Linux platform and investigate what kind of applications they are best suited for. Another purpose is to analyse what strengths/weaknesses they have compared to each other. The application types that we used were processor, memory, hard drive and network intensive applications. Four applications with different characteristics were installed on two clusters based on different process migrating systems. The applications were run on the clusters and interesting values were collected and interpreted. The process migrating systems of today are not yet fully developed and some functionality is missing. We believe that the process migrating systems have the future ahead of them. One of the process migration systems that we used is usable today, and the other one have been improved since we started to work on this thesis.