Cluster Infrastructure & Computing Building Local Cluster Environments

Abstract

Abstract The researcher aimed through this thesis to investigate the parallel computing environment in order to enhance performance and achieve maximum efficient usage of computer hardware to support scientific research and scientific experiments the computational unit, School of Physics and applied physics in Faculty of Science and Technology at Alncelain University. The aim was to build a low cost computer environment based on available PCs to stand for the costly and unavailable large mainframe computers due to technological restrictions on our country. Through this thesis the researcher studied the methodology of creating computer clusters and their testing means. Furthermore, a possible extendable and incremental cluster of four nodes was created using Intel Pentium 4 processor. Performance tests were run using a standard benchmark developed by NASA and Ohio State University namely NPB3.3 and OSU Micro-Benchmarks 3.5.1 respectively. Message Passing Interface (MPI) Matrix Multiplication A*B = C Parallel benchmark code, was implemented to capture latency which was found to range from 48.95 to 844447.34 micro second and Bandwidth which was found to range 0.01 to 9.96 MB/s.