Prediction of communication performance for wide area computing systems

Abstract

This paper presents a new algorithm to predict the communication performance for high performance parallel computing in wide area networks. Such a powerful prediction allows to balance the requirements of communication time and computation time. Such a system can be used in almost any wide area computing application. The major feature of this algorithm is the consideration of the human rhythm of life. Prediction values are calculated using a selected number of arguments, depending on daytime and weekday. Thus it is possible to give very accurate prediction values several weeks ahead. In this paper we discuss the usage of this method in the hypercomputer at the University of Restock. The prediction algorithm and monitoring tools are written in Java and can thus be used on almost any platform.