PERFORMANCE EVALUATION OF LARGER MATRICES OVER CLUSTER OF FOUR NODES USING MPI
Shambhavi Balamuthu Sampath, Bharat Bhushan Sagar · 2014
Parallel computing operates on the principle that large problems can often be divided into smaller ones, which are then solved concurrently to save time (wall clock time) by taking advantage of non-local resources and overcoming memory constraints. The main aim is to form a cluster based parallel computing architecture for MPI based applications which demonstrates the performance gain and losses achieved through parallel processing using MPI. This can be realized by implementing the parallel applications like solving matrix multiplication problem, using MPI. The architecture for demonstrating MPI based parallel applications works on the Master-Slave computing paradigm. We aim to evaluate the time statistics of parallel execution and do comparison with the time taken to solve the same problem in serial execution. We also demonstrate communication overhead involved in parallel computation. The results with runs on different number of nodes are compared to evaluate the efficiency of MPI based parallel applications. We also show the performance dependency of parallel and serial computation, on RAM. Finally we show the relationship between the number of slave processes to be specified for computation and the number of cores available for parallel computation.