Biomedical Engineering Reference
In-Depth Information
CHAPTER 7
Special-Purpose Computing for
Biological Sequence Analysis
BERTIL SCHMIDT and HEIKO SCHRÖDER
We present a hybrid parallel system based on commodity components to gain super-
computer power at low cost. The architecture is built around a coarse-grained PC
cluster linked to a high-speed network and fine-grained parallel processor arrays
connected to each node. Identifying applications that profit from this kind of com-
puting power is crucial to justify the use of such a system. In this chapter, we
present an approach to high-performance protein database scanning with hybrid
computing. To derive an efficient mapping onto this architecture, we have designed
instruction systolic array implementations for the Smith-Waterman and Viterbi algo-
rithm. This results in a database scanning implementation with significant runtime
savings.
7.1 INTRODUCTION
SIMD architectures looked very promising around 1990, as they delivered high-
est performance at the price of being special purpose engines rather than general
purpose. Then, with the end of the cold war, the western military stopped the
support for such systems. In 1998, the last SIMD supercomputer vanished from
the list of the top 500 most powerful supercomputers (see http:
//
www.top500.org
/
lists
/
2004
/
06
/
overtime.php?c=1). What was it that made the SIMD concept com-
pelling in first place and what caused the end of this concept? Advantages of the
SIMD concept over the MIMD concept include the following aspects:
Massively parallel SIMD computers typically consist of at least 2
10
processing
elements, but only one control unit. This single control unit broadcasts a single
•
Search WWH ::
Custom Search