Grid Scheduling Optimization Under Conditions of Uncertainty

One of the biggest challenges in building grid schedulers is how to deal with the uncertainty in what future computational resources will be available. Current techniques for Grid scheduling rarely account for resources whose performance, reliability, and cost vary with time simultaneously. In this paper we address the problem of delivering a deadline based scheduling in a dynamic and uncertain environment represented by dynamic Bayesian network based stochastic resource model. The genetic algorithm is used to find the optimal and robust solutions so that the highest probability of satisfying the user’s QoS objectives at a specified deadline can be achieved. It is shown via a simulation that the new methodology will not only achieving a relatively high probability of scheduling workflow with multiple goals successfully, but also be resilient to environment changes.