Parallel and pipelined digital signal processor system for fast protein sequence homologies

Abstract

A study to evaluate the architecture of a multi-DSP (digital signal processing) system primarily intended to execute, at high speed, a very sensitive algorithm that finds homologies between protein sequences is presented. This algorithm is an ideal candidate for implementation in a multiprocessing environment using parallel and pipelined processing. This study of a multiple-DSP system for finding protein sequence homologies against large databases showed that, even with the sophisticated processors available today, some extra work has to be added either in hardware by building powerful input/output components or in the coding of the software. The critical parts in number crunching algorithms still deserve to be looked at carefully and it is probably justified to program such parts at the assembly level to get the best possible speed out of the system.<>