A Comparative Analysis of the Smith-Waterman Algorithm on Raspberry Pi-Based Parallel Platforms

Sequence alignment is a fundamental task in bioinformatics, requiring intensive computational processing that grows quadratically with sequence size. High-Performance Computing (HPC) offers essential solutions to accelerate such tasks. This paper presents a detailed performance analysis of the local alignment Smith-Waterman algorithm, comparing a sequential implementation against parallel versions designed for modern multi-core and nodes architectures. For shared-memory parallelism, an OpenMP version was developed using a wave front strategy to manage data dependencies; for distributed-memory, an MPI version was implemented using a 2D row-based domain decomposition. The evaluation results, using workloads of sequences sizes of 1000, 5000, and 15000, revealed distinct performances. The OpenMP approach proved effective for larger workloads (peak speedup of 1.84x), though inefficient for small workloads (speedup of 0.56x) due to parallelization overhead. In contrast, the MPI approach was consistently outperformed by the sequential version in all tests, demonstrating that the high cost of inter-node communication nullified the gains from distributed computing. The analysis concludes that the choice of a parallel model must carefully balance architectural paradigms with algorithmic characteristics to achieve meaningful performance improvements.