Authors
Do Hyun Bae School of Computer Sciences, University of Seoul, Seoul, Korea ilsoo99@uos.ac.kr
Munkhbayar Bayartsogt School of Computer Sciences, University of Seoul, Seoul, Korea muba@uos.ac.kr
Jin Suk Kim School of Computer Sciences, University of Seoul, Seoul, Korea kimjs@uos.ac.kr
In this paper we introduce a parallel approach to calculate massive matrix inversion. It needs large size of memory to compute with large size of matrices. Because of the memory requirements, we consider an algorithm to optimize memory distribution in cloud computing system. In matrix inversion using Gauss-Jordan algorithm, we found out a lot of regional memory access tendency in the algorithm. We also consider this memory access tendency. To solve these problems, we divide the matrix data as many as numbers of processors which was assigned to calculate matrix inversion. Dividing directions both horizontal and vertical are possible to imply. Matrix inversion has steps, and this step is increase according to the size of the matrix, and previous step calculation results are used at each step calculation results. To do above process, we use a parallel scheduler. Parallel scheduler manages the all processors and synchronizes these processors calculation. Research is focused on solving massive matrix inversion, so we test our research in cloud computing system, and we obtain the progress results.
Do Hyun Bae School of Computer Sciences, University of Seoul, Seoul, Korea ilsoo99@uos.ac.kr
Munkhbayar Bayartsogt School of Computer Sciences, University of Seoul, Seoul, Korea muba@uos.ac.kr
Jin Suk Kim School of Computer Sciences, University of Seoul, Seoul, Korea kimjs@uos.ac.kr
In this paper we introduce a parallel approach to calculate massive matrix inversion. It needs large size of memory to compute with large size of matrices. Because of the memory requirements, we consider an algorithm to optimize memory distribution in cloud computing system. In matrix inversion using Gauss-Jordan algorithm, we found out a lot of regional memory access tendency in the algorithm. We also consider this memory access tendency. To solve these problems, we divide the matrix data as many as numbers of processors which was assigned to calculate matrix inversion. Dividing directions both horizontal and vertical are possible to imply. Matrix inversion has steps, and this step is increase according to the size of the matrix, and previous step calculation results are used at each step calculation results. To do above process, we use a parallel scheduler. Parallel scheduler manages the all processors and synchronizes these processors calculation. Research is focused on solving massive matrix inversion, so we test our research in cloud computing system, and we obtain the progress results.
0 Сэтгэгдэл:
Post a Comment