Parallel BLAST Analysis and Performance Evaluation

BLAST is an efficient heuristic algorithm used for comparing biological sequences, such as amino-acids or nucleotides and identifying similarities between them. However, extensive genetic mapping projects feed sequence databases continuously, forcing them to grow steadily, turning execution time into a potential constraint to sequential implementations of the algorithm and driving the search for new approaches, such as parallelism, to reduce it. In this work, parallel implementations of BLAST are evaluated, in particular mpiBLAST, aiming to find the best conditions for executing those applications in high performance parallel systems. Through comparative tests and a time profile of the application, we analyzed its performance variation regarding parameters such as load balancing, advanced load of database fragments in RAM and also employing a high throughput parallel file system.