Dynamic Deployment of Custom Execution Environments in Grids

One of the most important obstacles when porting an application to the Grid is its highly heterogeneous nature. This heterogeneity usually means an increase of the cost of both the application porting cycle and the operational cost of the infrastructure. Moreover, the effective number of resources available to a user are also limited by this heterogeneity. In this paper we presents two approaches to tackle these problems: (i) an straightforward deployment of custom virtual machines to support the application execution; (ii) and a new architecture to provision computing elements that allows to dynamically adapt them to changing VO demands. Experimental results for both approaches on prototyped testbed are discussed. In particular, the on-demand provision of computing elements show less than a 11% overall performance loss including the hypervisor overhead.