Comparison on efficiency of computational efforts between cluster computation (MapReduce) and single host computation

Abstract

The complexities of research in science have been increasing extremely. Numerous mathematical models have been developed. Matrix has been used popularly to model numerous and complex science and engineering problems. It is found that as the dimension of the matrix grows in size, the complexities of matrix computation increase. This problem may be solved by using large computer system (i.e. mainframe). However, its operational is very costly. Another solution is to utilize parallel computing, which are able to cut off the operational cost. A recent advance in parallel programming is the introduction of MapReduce, as a new approach in parallel programming. MapReduce can perform calculations with distributed method by utilizing an idle processor. In this research, the performance of MapReduce in matrix operation is compared to other conventional methods, which are Single Processor and Threads. The performances are assessed by comparing the execution time, CPU usage, and RAM usage of each approach. The results show that MapReduce performed better than the other approaches.