Performance Analysis of Algorithms on Shared Memory, Message passing and Hybrid Models for Standalone and Clustered SMPs

While algorithms are well-understood in its sequential form, comparatively little would be known of how to implement parallel algorithms with main-stream parallel programming platforms and run it on SMP-based mainstream systems such as multi-core clusters. The project aims at better understanding the algorithmic techniques like divide and conquer, decrease and conquer, transform and conquer parallel algorithms on parallel platforms. In this paper we investigate four benchmark parallel algorithms such as Quick sort, Matrix Multiplication, Montecarlo, LU Decomposition on Shared memory algorithms that run on SMPbased mainstream systems with OpenMP, Message-passing model on computer clusters with Open MPI, and combined Hybrid algorithms with combination of OpenMP and MPI that run on clustered SMPs. Parallel Benchmark programs are run on a dedicated Beowulf cluster and their performance were investigated.