期刊名称：International Journal of Advanced Research In Computer Science and Software Engineering
印刷版ISSN：2277-6451
电子版ISSN：2277-128X
出版年度：2013
卷号：3
期号：4
出版社：S.S. Mishra
摘要：This paper elucidates the implementation and performance evaluation of an "off the shelf" cluster. A distributed memory cluster was designed using "dumped" heterogeneous machines and the gain in performance was measured for various configurations. The idea has been coined as junk computing. Creating a faster computer utilizing existing computers is an idea that has been long studied and widely researched. We havei attempted to add a new dimension to this study by running several programs on our cluster made up of 3 ancient Pentiums and a young i3 processor, and comparing the performance gain achieved under various configurations. Of special interest is the case where the performance of the three-node cluster of Pentium 4 is compared with the individual performance of an i3-330M processor. The gain achieved for linear problem is more in the cluster made of two P4 processors than the gain achieved in a single i3 processor. It is also evident from the research results that two Pentium 4 processors can compete with younger generation processor like Intel i3-330M. The study also confirms the intuitive and known notions of the parallel computing, so we would also bring about some implications of the obvious. For example, it was noticed that distributing the computations over the cluster yields gain only after a certain threshold that is after the point when the communication costs become negligible as compared to computatio n cost. It was also noticed that we might not gain much by adding a relatively slower node to the cluster. Before concluding the paper, we examine a shortcoming of the cluster; the case of distributed memories whereby a performance hit of the factor of up to 20 was observed
关键词：Cluster; Intel(R) Pentium(R); E waste management; Amdahl's Law; mpih2
