Dynamic Provisioning of Virtual Clusters for Grid Computing

Virtual machines can greatly simplify grid computing by providing an isolated, well-known environment, while increasing security. Also, they can be used as the base technology to dynamically modify the computing elements of a grid, so providing an adaptive environment. In this paper we present a Grid architecture that allows to dynamically adapt the underlying hardware infrastructure to changing Virtual Organization (VO) demands. The backend of the system is able to provide on-demand virtual worker nodes to existing clusters and integrate them in any Globus-based Grid. In this way, we establish a basis to deploy self-adaptive Grids, which can support different VOs in shared physical infrastructures and dynamically adapt its software configuration. Experimental results on a prototyped testbed show less than a 10% overall performance loss including the hypervisor overhead.