Predicting the Performance of Data Transfer in a Grid Environment

In a Grid environment, implementing a parallel algorithm for data transfer or multiple parallel jobs allocation doesn’t give reliable data transfer. There is a need to predict the data transfer performance before allocating the parallel processes on grid nodes. In this paper we propose a predictive framework for performing efficient data transfer. Our framework considers different phases for providing information about efficient and reliable participating nodes in a computational Grid environment. Experimental results reveal that multivariable predictors provide better accuracy compared to univariable predictors. We observe that the Neural Network prediction technique provides better prediction accuracy compared to the Multiple Linear Regression and Decision Regression. Proposed ranking factor overcomes the problem of considering fresh participating nodes in data transfer.