On the Feasibility of Using Network Processors for DNA Queries

This chapter demonstrates that NPUs are a promising platform for implementing certain algorithms in bio-informatics that could benefit from the NPU's parallelism. is the chapter shows how the Blast algorithm is successfully implemented on a fairly slow IXP NPU. The performance is equivalent to an implementation on a Pentium running at a much higher clock rate. A few attempts were made to optimize the solution. Although the Aho-Corasick algorithm itself is fairly efficient, many improvements of Blast have been proposed. Moreover, several simple changes to IXPBlast are expected to boost performance significantly. First, a trivial improvement for IXPBlast is to switch to amino acids (effectively looking at three nucleotides at a time). Second, the Aho-Corasick algorithm that was used in IXPBlast can itself be improved. Research projects exist that make the algorithm either faster or more memory efficient. Third, while implementing the feedback mechanism between sender and IXP will not change the number of cycles spent on processing the data, it will decrease the duration of the process as experienced by the user. Fourth, multiple network processors can be used in parallel by plugging in multiple ENP NPU boards in the PCI slots of a PC to speed up the process even more.