Single System Image clustering using Kerrighed

Abstract

In a compute cluster, the nodes may be distributed and heterogeneous in nature. It may be difficult to administer and interface with such a system. It will be better if all the users get an interface with same look and feel rather to have a different view in each node of the cluster. Thus Single System Image(SSI) is a major design goal of cluster based systems. The services that SSI clusters offer are single user interface, single process space, single memory space, single file hierarchy and single control point. Kerrighed is an SSI Operating System which provides the services mentioned above. This paper gives the details about how to set up a Kerrighed cluster, how to use a kerrighed cluster for running MPI programs, and shows how a cluster can improve the performance of an MPI program when it is run on a cluster of machines. The paper also gives an idea about how Kerrighed helps programmers to easily add new scheduling algorithms for cluster with the help of Scheduler API and a design for adding a proposed algorithm into Kerrighed.