Parallel merge sort based performance evaluation and comparison of MPI and PVM

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. Merge sort is not only a fast and stable sort algorithm, but it is also an easy to understand and popular representative of the rich class of divide-and-conquer methods, hence better understanding of merge sort parallelization can contribute to better understanding of divide-and-conquer parallelization in general. The main aim is to form a common single node model for both MPI and PVM which demonstrates the performance dependency of parallel merge sort on RAM of the nodes (desktop PCs) used in parallel computing. The single node model consists of a user, a master, capable of handling requests from the user and a slave, capable of accepting problems from the master and sending the solution back. The master and the slave communicate with each other using MPICH-2. The master will monitor the progress and be able to compute and report the time taken to solve the problem, taking into account the time spent in assigning the problem into slave and sending the results along with the communication delays. We aim to evaluate and compare these statistics to decide which among MPI and PVM gives faster performance. And also we compare with the time taken to solve the same problem in serial execution to demonstrate communication overhead involved in parallel computation. We aim to compare and evaluate the time statistics obtained for different sizes of RAM under parallel execution using MPI and PVM in a single node involving only two cores, where one acts as master and other as slave. We also show the dependency of serial execution on RAM for the same problem by executing its serial version under different sizes of RAM.