NCBI BLASTP on the convey HC1-EX

The BLAST sequence alignment program is a central application in bioinformatics. The de facto standard version, NCBI BLAST, uses complex heuristics which make it challenging to simultaneously achieve both high performance and exact agreement. In previous work, a system that used novel FPGA-based filters reduced the input database by over 99.97% without loss of sensitivity. In the present work we report experiences in getting from a prototype to a potential product for the Convey HC1-EX. There are several issues. The first is the efforts made to maintain timing for a highly complex configuration as it is optimized by including additional filter stages. This requires implementation and optimization of new interface logic as well as floor-planning. The second is the system-level tradeoffs necessary to maintain correctness. The issue here is preventing low frequency events, which necessarily cannot be mapped to the FPGA, from diluting the performance benefits without sacrificing sensitivity. We present results for various usage scenarios and find a factor of nearly 5x speed-up over a fully parallel implementation of the reference code on a contemporaneous CPU. We believe that the resulting system is the leading accelerated NCBI BLAST. The significance of this work is that, while such in-depth work is necessary to achieve highperformance for complex systems, these issues are rarely described nin the academic literature.