Performance modelling of parallel BLAST using Intel and PGI compilers on an infiniband-based HPC cluster

Area of Science:

• Bioinformatics
• Computational Biology
• High-Performance Computing (HPC)

Background:

• The Basic Local Alignment Search (BLAST) is a crucial bioinformatics tool for sequence similarity searches.
• Scaling BLAST to large genomic databases presents computational challenges.
• Optimizing BLAST performance on HPC systems is essential for modern biological research.

Purpose of the Study:

• To map and evaluate serial and parallel BLAST algorithms on an Infiniband-based High Performance Cluster.
• To assess the scalability and efficiency of BLAST implementations using different parallel compilers.
• To identify optimal configurations for efficient BLAST execution on state-of-the-art HPC systems.

Main Methods:

• Implementation and evaluation of serial and parallel BLAST algorithms.
• Performance benchmarking on a large Infiniband-based HPC system.
• Comparative analysis using Intel and Portland's PGI parallel compilers.

Main Results:

• Demonstrated runtime scalability of the BLAST algorithm on the HPC system.
• Achieved up to 87% efficiency for parallel BLAST execution.
• Identified the impact of MPI suite, compiler, interconnect, and CPU on performance.

Conclusions:

• Effective scalability of BLAST runtime is achievable on HPC systems.
• The right combination of software and hardware components is key to maximizing BLAST efficiency.
• This work provides valuable insights for optimizing large-scale sequence analysis in bioinformatics.