A Markov Chain Based Resource Prediction in Computational Grid

Due to the dynamicity of resources and owners' behaviors, the grid resource state changes continuously. As a result, scheduling strategies only based on current static information can no longer meet the needy of more effective application. Prediction of future resource state which combines current state and historical records can improve the efficiency and reliability of scheduling. In this paper, we present a Markov chain bared prediction method, which comprehensively considers the rate of CPU usage, level of network load, and resource failure rate to forecast resource future state for getting better job scheduling results. An evaluation measurement is designed to quantify, the prediction result for scheduling. Experiments show that this method has a better performance on resource idle rate and prediction accuracy.