On a parallel algorithm for the determination of multiple optimal solutions for the LCSS problem

Abstract

For particular real world combinatorial optimization problems e.g. the longest common subsequence problem (LCSSP) from Bioinformatics, determining multiple optimal solutions (DMOS) is quite useful for experts. However, for large size problems, this may be too time consuming, thus the resort to parallel computing. We address here the parallelization of an algorithm for DMOS for the LCSSP. Considering the dynamic programming algorithm solving it, we derive a generic algorithm for DMOS (A-DMOS). Since the latter is a non perfect DO-loop nest, we adopt a three-step approach. The first consists in transforming the A-DMOS into a perfect nest. The second consists in choosing the granularity and the third carries out a dependency analysis in order to determine the type of each loop i.e. either parallel or serial. The practical performances of our approach are evaluated through experimentations achieved on input benchmarks and random DNA sequences and targeting a parallel multicore machine.