Comparison of a cost-effective virtual cloud cluster with an existing campus cluster

Abstract

The Cloud provides impartial access to computer services on a pay-per-use basis, a fact that has encouraged many researchers to adopt the Cloud for the processing of large computational jobs and data storage. It has been used in the past for single research endeavours or as a mechanism for coping with excessive load on conventional computational resources (clusters). In this paper we investigate, through the use of simulation, the applicability of running an entire computer cluster on the Cloud. We investigate a number of policy decisions which can be applied to such a virtual cluster to reduce the running cost and the effect these policies have on the users of the cluster. We go further to compare the cost of running the same workload both on the Cloud and on an existing campus cluster of non-dedicated resources.