Anticipating resource saturation in Federated Grids

In a dynamic and geographically distributed Federated Grid where resources are shared between system participants, there is a lack of mechanisms capable of reallocating already scheduled tasks based on grid infrastructure owners' current internal needs. In this paper we propose a set of policies for both, users and owners, that aid owners to satisfy internal peak demands and users to achieve the best makespan despite the circumstances. As in our previous work, the main purpose is to do so in the least intrusive way possible to maintain software stack independence of all participants, and to save time and communication bandwidth by anticipating grid resources saturation. These strategies suppose a novel approach for decentralized and non-cooperative workflow scheduling in a federation of heterogeneous grid infrastructures. We evaluate and prove the feasibility of our policies through a set of simulations that reflect the worst case where all resources are saturated. The results show that, in the worst scenario, our scheduling mechanism is beneficial to big infrastructure owners since they can achieve their own internal objectives, as well as to small users since they can reach the best possible completion time.