Performance Evaluation of Network based Distributed Supercomputing Environment

Abstract

In the past decade, supercomputing has witnessed a paradigm shift from massively parallel supercomputers to network computers. Though dedicated high end supercomputers still have their place in the market yet combined unused CPU cycles of desktop PCs available in the campus network can form comparable virtual supercomputers. Consequently, Parallel Processing in a network of PCs are attracted a boost of attention and becoming one of the most promising areas of large scale scientific computing. In this paper, we are presenting Grid-enabled PC Cluster (GPCC), exhibiting low latency and bandwidth scalable sub-communication system. The design of the GPCC is such that it keeps in view the socket buffer size of local and non-local nodes in the network environment. The design is relatively easy to use, inexpensive to apply and extremely accurate. The highly accurate results provided by TCP/IP ping-pong were coupled with parallel matrix multiplication benchmark. Parallel
Matrix Multiplication (PMM) performance benchmark is used to test the GPCC for node-to-node network performance and parallel floating point performance of all involved processor in a local and non-local cluster environment. PMM benchmark is developed on the basis of master-slave model using dynamic distribution scheme.