Comparison of Acceleration Techniques for Selected Low-Level Bioinformatics Operations.

Area of Science:

• High-Performance Computing (HPC)
• Bioinformatics
• Computational Life Sciences

Background:

• Stagnating processor clock rates necessitate alternative strategies for increasing computing power.
• Rapid data growth in life sciences and bioinformatics demands efficient data processing solutions.
• Increased multi-core processors highlight the importance of parallelization in software development.

Purpose of the Study:

• To provide an overview of manual and automatic acceleration techniques for high-performance computing.
• To analyze the performance and applicability of automatic acceleration approaches in sequence informatics.
• To guide readers in selecting appropriate acceleration techniques for their specific computational challenges.

Main Methods:

• Review of state-of-the-art manual and automatic acceleration techniques.
• Comparative analysis of four automatic acceleration approaches: OpenMP, PluTo-SICA, PPCG, and OpenACC.
• Demonstration of automatic acceleration for two use cases (k-mer analysis and matrix multiplication) to illustrate challenges and benefits.

Main Results:

• Automatic code parallelization is feasible with current compiler technology, yielding significant execution speed improvements.
• CPU-targeted automatic optimizers (e.g., OpenMP) are mature and often require no manual adjustments.
• GPU-targeted automatic parallelizers show promise but may lack maturity, being limited to simpler code structures; performance gains are problem-size dependent due to data migration overhead.

Conclusions:

• Automatic code parallelization offers a viable solution for accelerating scientific applications.
• CPU optimization tools are generally mature, while GPU optimization tools require further development for broader applicability.
• The choice between CPU and GPU optimization depends on the specific use case, problem size, and data characteristics.