Skip to Main Content
For the heterogeneous and dynamic nature of grid environments, the ability to accurately and timely predict future capabilities of resources is very important. Autoregressive models are appropriate and much less expensive for predicting host load, but Autoregressive modeling includes a model identification procedure, that is, it is necessary to choose the order that best describes the host load variety. In this paper four of suggested criteria to determine the optimal order of AR models have been evaluated: The final prediction error (FPE), Akaike's information criterion (AIC), minimum description length (MDL) and the Bayesian information criterion (BIC). We evaluated these criteria on four of long, fine grain load traces from a variety of real machines, and our experimental results demonstrate that BIC criteria has the best determination of the optimal order than others and the optimal orders of AR models should be different in heterogeneous machines for load prediction.