Performance evaluation of predictive replica selection using neural network approaches

The ability to accurately predict a best replica from different sites holding replicas of a particular file is of great importance for applications that require access to replicated files for their execution. The best replica is the one that optimizes the desired performance criterion such as speed, cost, security or transfer time. As grid is dynamic in nature, the predicted best site for replica selection may not be the best site for replica selection with current network conditions. Neural network approaches address such dynamism and predict the best site more accurately for change in the network conditions. In this paper we compare and evaluate the prediction differences of various neural network approaches for replica selection problem.