Grid resource prediction approach based on Nu-Support Vector Regression

One of the challenging problems in grid environment is the choice of destination nodes where the tasks of the application are to be executed. Therefore, resource prediction is a crucial direction for job scheduling system and grid users. In this paper, Nu-support vector regression (v-SVR) is applied to solve resource prediction problem. The method of parallel multidimensional step search is also introduced to select parameters for v-SVR prediction model. Standard v-SVR method is extended to a systematic approach for grid developer and user to finish preprocessing of data set and model optimization with high prediction accuracy. Experiments with resource data set were performed on computing nodes in grid environment. Statistical analysis shows that our approach can automatically locate suitable parameters for building prediction model with high accuracy and remarkably reduce the computational time of model optimization.