A Flexible Architecture for Scheduling Parallel Applications on Opportunistic Computer Networks?

Opportunistic computer networks are a promising environment to execute parallel ap-plications due to the processing power supplied by the current desktop machines. Parame-ter sweep and master-worker applications are suitable for these environments since a fail-ure of a process does not necessarily suspend the overall application. On the other hand, in tightly-coupled parallel applications a failure in a machine may interrupt or crash the entire application. Thus, different application classes require different scheduling strate-gies. Furthermore, in the context of Computational Grids, each Grid domain has its own administrative policies, user and application profiles and resource properties. All these reasons have motivated us to propose a flexible and extensible architecture for schedul-ing parallel applications which relies on different services designed as plug-ins. We also discuss the integration of this scheduling architecture in the InteGrade project which is a novel middleware infrastructure for Grid Computing.