Preliminary Analysis of the Cell BE Processor Limitations for Sequence Alignment Applications

Multicore Architectures

“…It is freely available and there are web servers [4,5] to let users run database searches using the FASTA tools. FASTA takes a given nucleotide or aminoacid sequence and searches a corresponding sequence database by using local sequence alignment to find matches of similar database sequences.…”

“…The main performance bottleneck is the long computation time needed for each PW alignment because of the use of sequential codes running on a rather old processor (Pentium III). More recently, other works [52,53,80,93] have studied ClustalW's performance on the Cell BE [55]. In [93] ClustalW is parallelized for Cell BE.…”

“…However, the traditional focus on multimedia and networking applications have not allowed bioinformatics to efficiently benefit from current architectures. In [52], we investigated the limitations of the Cell BE architecture when targeting bioinformatics. Some of the limitations mentioned there have been later considered by adding new instructions (Section 5.4).…”

“…Lastly, an even higher performance (688GCUPS) has been recently reported by a commercial FPGA-based product [14], likely using more hardware resources though. In Chapter 6 we study a multicore architecture where each core has the performance of either a general-purpose processor [43,44,52] or an application-specific accelerator [14,24,58,61]. Table 3.1 shows the difference in throughput of various implementations of the kernel targeted in Chapter 6.…”

“…The second phase in ClustalW (GT) has been shown [52,80,93] to take an almost negligible amount of time in most cases, certainly in the ones we are interested in, where sequences are in the order of a thousand symbols. Lastly, PA progressively aligns sequences following the tree structure.…”