Process speed-up by using parallel computers

PROCESS SPEED-UP BY USING PARALLEL COMPUTERS (M.Sc. Thesis) Muhammet ÜNAL GAZI UNIVERSITY INSTITUTE OF SCIENCE AND TECHNOLOGY December 2003 ABSTRACT Computing systems, which have been developed for saving time and work, cost high, and able to perform complex computations, are called "super computers". In this thesis: scalable computer systems, which are the fundamentals of super computers, have been examined in order to obtain this type of systems with less cost and more speed. The most suitable system is considered as "clusters". Studies indicate that Beowulf clusters have the lowest cost/performance ratio. Nodes of Beowulf clusters are composed of easily accessible PCs. Interconnection networks are composed of commodity Ethernet network hardware therefore forming home made supercomputer systems becomes possible. Through this vision and studies, the designed supercomputer system has 10 AMD Athlon XP 2800+ processors, 5 GByte Registered ECC memory and 1.2 TByte storage capacity. By using benchmarking programs the speed of the system and the speed-up capability of the processes have been measured. These measured values indicated that total average computing power of the system is 10 Gflop while one processor is 2 Gflop. With the obtained speed-up factor of 5, a powerful computer system has been established in order to make scientific computations and research. Science Code : 6190202 Key Words : Parallel Computers, Beowulf, Computer Clusters, Performance Analysis, Benchmarking Page Number: 605 Adviser : Assoc. Prof. Selma YÜNCÜ