A Nu-support Vector Regression Based System for Grid Resource Monitoring and Prediction

In order to realize intelligent scheduling of incoming tasks and provide acceptable quality of service, a distributed system for monitoring and prediction of computing grid resources and network conditions becomes inevitable. In this paper, we propose the design and implementation of computing grid resource monitoring and prediction system. The system is applicable in that it is robust, scalable, extensible, and user-friendly. Nu-support vector regression (Nu-SVR) is employed as modeling method of multi-step-ahead prediction, and a combinational optimization algorithm is proposed to jointly optimize feature selection and hyperparameter selection for prediction model. Performance evaluation on prediction methods is performed with benchmark data sets, whereas comparative results show that the Nu-SVR model has high prediction accuracy, and the combinational optimization algorithm can improve prediction performance efficiently; hence these two methods are suitable for online monitoring and prediction system.