An improved technique for program remodularization

Abstract

There has been a tremendous shift towards using clusters and networks of low-cost computers for high performance computing. High performance can be achieved by appropriate remodularization of computationally intensive programs. An appropriate modular structure of program exploits the inherent concurrency in the execution of the program. In this article a new relation to determine suitable modular structure for distributing programs across the network is presented. The relation is built automatically by analyzing the program call flow graph. The proposed relation outperforms the existing relation for computing the amount of concurrency in a distributed code by considering, situations in which more than one method are supposed to be executed on a single computational node. © 2010 Springer-Verlag.