Boosting data throughput for sequence database similarity searches on FPGAs using an adaptive buffering scheme

Searching on DNA and protein databases using sequence comparison algorithms has become one of the most powerful techniques to better understand the functionality of particular biological sequences. However, the requirements to process the biological data exceed the ability of general-purpose processors. FPGAs (Field Programmable Gate Arrays) connected to server processors have been used to accelerate similarity searches. However, reconfigurable computing platforms have utilized an external I/O bus as the communications channel, limiting the communication speed between the host processor and the FPGA. This communication bottleneck often offsets the application speedup enabled by FPGAs. In this paper we present an adaptive data prefetching scheme to avoid reconfigurable processing coprocessor stalls due to data unavailability through profiling methodologies and quantitative analysis. Experimental results on various query sequences show that the proposed scheme can effectively eliminate a major portion of the data access penalty, increase throughput of the FPGA implementation by up to 42%, and achieve a speedup of 110 for affine gap penalties over a standard PC implementation.