by

Communication is an important component in determining the overall performance of a distributed memory parallel computing application. It therefore becomes essential to predict communication performance of applications on the underlying network hardware of a target distributed memory system. This thesis concentrates on integrating a cycle driven k-ary n-cube network simulator to the existing CAL-SIM distributed memory simulator for evaluating communication performance of message passing applications. The design and implementation of a suitable network interface required for the integration is presented. With detailed network simulation the accuracy of predictions made is very high. The impact on communication performance by varying some of the network design parameters is studied. Other important aspect of this work is the access to an evaluation platform for evaluating network design tradeoffs, using real applications a