An empirical study of distribution based on Voyager: a performance analysis

The paper describes the model, implementation and experimental evaluation of a distributed Kohonen neural network application. The aim of this research is to empirically verify the suitability and the performance of a distributed application based on mobile objects and, in perspective, intelligent agents. This research aims to provide distribution features in decision support systems. The experiment was based in the Java-ABC project. The Java-ABC project is concerned with flexible software architecture for decision support systems which rely on artificial neural network (ANN) technology. Three parameters: used CPU, used memory (RAM) and time consumed by each Kohonen application, were taken as evaluation measures in this experiment. Three hardware environments were used: two PCs (one local and one remote) under Windows NT with different RAM capacity and a SUN Ultra under SunOS 5.6. This paper presents the comparison of performance measures from our experimental studies and the analysis of the results. In conclusion, the paper presents the implications of these results for the area of distributed intelligent decision support and future directions of this work.