Experimental Study of Parallelizing Breadth First Search (BFS) Algorithm

Recent years, many applications have exploited graph-based computations for searching massive data through distributed processors and memories. In this study, we present Breadth First Search algorithm as a graph-based algorithm to organize large DNA dataset and parallelize the algorithm using two highly tuned parallel approaches named OpenMP and MPI on a cluster, besides tuning the serial version of the algorithm. OpenMP is implemented using domain decomposition method. MPI is applied after dividing the dataset into equal parts and changing it to a two-dimensional array. Both approaches are tested in Khwarizmi cluster with 8 Intel Xeon Processors at the School of Computer Sciences in USM. The two aforementioned experiments are implemented and evaluated with certain characteristics and the results show high performance is achieved in terms of speedup and efficiency in comparison with the serial version of the algorithm.