Performance estimation of semirandom data transfer within direct hypercube interconnection network

A useful theoretical model for the transfer of data inside a hypercube multiprocessing network is developed. This model defines a statistical parameters lambda which can be interpreted both as an average distance travelled by data within the network and as a measurement of how much memory sharing occurs within the network during the execution of a program. A simple model for the node in a distributed memory hypercube is derived to show data flow. Methods of estimating the time required for data transfer are also developed. A simulation for the data flow model is included.