Improved Approach for Unbalanced Load-Division Operations Implementation on Hybrid Parallel Processing Systems

Abstract

The modern computer-systems designed according to multiprocessor configurations. Multiple processors enable multiple threads to be executed simultaneously with the ability of executing the threads of the same process to be run on different processors at the same time. This paper addresses the building of a software application to be implemented on hybrid memory systems depending on client/server principles, the network can contain any number of nodes; one of them is a client and the others are servers. An improved approach was produced for problem subdivision based on an unbalanced load division case study (Matrix multiplication). Many previous drawbacks overcame, such as matrix-size limitation, effect of multi-core with distributed systems and forcing the processes and threads among multi-core system processors. Thus, the communication-direction from client-side toward the servers-side and vice-versa became more powerful by binding the activities of both Massage-Passing-Interface (MPI) with those of Open Multi-Processing (OpenMP). The proposed algorithms are executed by Quasar Toolkit (QT) creator application using C++ and QT library. The application-software is implemented to get high speed with as possible as minimum time and detect the effects of this system on the CPU Execution time and CPU Usage, the results are very acceptable and the processing time is decreased by 5.4492 times comparing with those without using hybrid parallel processing.