Improving the Mapping of Smith-Waterman Sequence Database Searches onto CUDA-Enabled GPUs

Area of Science:

• Bioinformatics
• Computational Biology
• High-Performance Computing

Background:

• Sequence alignment is fundamental to bioinformatics.
• The Smith-Waterman algorithm is a key tool for sequence searching.
• Mapping algorithms to graphics processing units (GPUs) offers computational advantages.

Purpose of the Study:

• To improve the performance of the Smith-Waterman algorithm on GPUs, particularly for short query sequences.
• To optimize the utilization of shared memory in GPU implementations.
• To analyze the impact of thread and block allocation on performance.

Main Methods:

• Developed a novel mapping strategy for the Smith-Waterman algorithm on GPUs.
• Focused on enhanced shared memory allocation for short query sequences.
• Evaluated the method on Tesla C1060 and Tesla K20 GPUs, comparing against CUDASW++.
• Analyzed performance variations with different thread and block configurations.

Main Results:

• The proposed method demonstrates significant performance improvements for the Smith-Waterman algorithm on CUDA-enabled GPUs.
• Optimized shared memory usage leads to better efficiency, especially for short sequences.
• Proper allocation of blocks and threads is crucial for maximizing performance gains.

Conclusions:

• The enhanced Smith-Waterman algorithm mapping provides a substantial speedup on GPUs.
• Shared memory optimization is a key factor in improving GPU-based sequence alignment.
• This work contributes to more efficient bioinformatics analyses through hardware acceleration.