21.01.2013 Views

Lecture Notes in Computer Science 4917

Lecture Notes in Computer Science 4917

Lecture Notes in Computer Science 4917

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Drug Design Issues on the Cell BE<br />

Harald Servat 1 , Cecilia González-Alvarez 2 , Xavier Aguilar 1 ,<br />

Daniel Cabrera-Benitez 2 , and Daniel Jiménez-González 2<br />

1 Barcelona Supercomput<strong>in</strong>g Center<br />

2 Universitat Politècnica de Catalunya<br />

Jordi Girona 1-3, Campus Nord-UPC, Modul C6, E-08034 Barcelona<br />

harald.servat@bsc.es xavier.aguilar@bsc.es,<br />

cecilia,dcabrera,djimenez@ac.upc.es<br />

Abstract. Structure alignment prediction between prote<strong>in</strong>s (prote<strong>in</strong><br />

dock<strong>in</strong>g) is crucial for drug design, and a challeng<strong>in</strong>g problem for bio<strong>in</strong>formatics,<br />

pharmaceutics, and current and future processors due to it<br />

is a very time consum<strong>in</strong>g process. Here, we analyze a well known prote<strong>in</strong><br />

dock<strong>in</strong>g application <strong>in</strong> the Bio<strong>in</strong>formatic field, Fourier Transform<br />

Dock<strong>in</strong>g (FTDock), on a 3.2GHz Cell Broadband Eng<strong>in</strong>e (BE) processor.<br />

FTDock is a geometry complementary approximation of the prote<strong>in</strong><br />

dock<strong>in</strong>g problem, and basel<strong>in</strong>e of several prote<strong>in</strong> dock<strong>in</strong>g algorithms<br />

currently used. In particular, we measure the performance impact of reduc<strong>in</strong>g,<br />

tun<strong>in</strong>g and overlapp<strong>in</strong>g memory accesses, and the efficiency of<br />

different parallelization strategies (SIMD, MPI, OpenMP, etc.) on port<strong>in</strong>g<br />

that biomedical application to the Cell BE. Results show the potential<br />

of the Cell BE processor for drug design applications, but also that<br />

there are important memory and computer architecture aspects that<br />

should be considered.<br />

1 Introduction<br />

Prote<strong>in</strong>-prote<strong>in</strong> dock<strong>in</strong>g algorithms predict the structure alignment between two<br />

or more prote<strong>in</strong>s to form a complex, without the need of experimental measurements<br />

at the laboratory. The computational cost of those algorithms is high<br />

due to the large number of aspects to consider. There are three ma<strong>in</strong> types of<br />

dock<strong>in</strong>g: flexible dock<strong>in</strong>g, rigid-body dock<strong>in</strong>g, and their comb<strong>in</strong>ation. In this paper<br />

we focus on the rigid-body dock<strong>in</strong>g, that consideres that the geometries of<br />

the prote<strong>in</strong>s are not modified when form<strong>in</strong>g the complex. Rigid-body dock<strong>in</strong>g is<br />

<strong>in</strong>adequate when there are several conformational changes dur<strong>in</strong>g the complex<br />

formation, but it is a very good basel<strong>in</strong>e for flexible prote<strong>in</strong> dock<strong>in</strong>g algorithms.<br />

In particular, we focus on the f<strong>in</strong>e- and coarse-gra<strong>in</strong> parallelization of the<br />

Fourier Transform Dock<strong>in</strong>g (FTDock) application [14,20] on a Cell processor.<br />

We present experiment results us<strong>in</strong>g MPI, OpenMP, function offload<strong>in</strong>g, and<br />

vectorization. Note however that we do not <strong>in</strong>tend to propose the fastest version<br />

of the FTDock on a Cell BE blade. Our ma<strong>in</strong> objective is to show the performance<br />

impact of different implementation and parallelization decisions that one<br />

can take when port<strong>in</strong>g code to a Cell BE blade.<br />

P. Stenström et al. (Eds.): HiPEAC 2008, LNCS <strong>4917</strong>, pp. 176–190, 2008.<br />

c○ Spr<strong>in</strong>ger-Verlag Berl<strong>in</strong> Heidelberg 2008

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!