More Bio-HPC Highlights From SC09

By Matthew Dublin

The best paper award at this year's SC09 went to the folks at D.E. Shaw Research, for their work entitled "Millisecond-Scale Molecular Dynamics Simulation on Anton." The much-touted Anton is a recently completed, specially designed supercomputer for molecular dynamics simulations that uses a huge number of interconnected ASICs. Anton allows researchers, for the first time, to simulate biological molecules at atomic-level detail for periods on the order of a millisecond. In their paper, they presented Anton's performance of MD simulations that had been validated against existing MD software and experimental observations.

A trio of researchers from the University of Tennessee, Knoxville introduced attendees to a few of their highly-scalable bioinformatics tools tweaked expressly for use on supercomputers. The team, which includes UTK joint faculty professor Igor Jouline and NICS computational scientist Christian Halloy, presented their versions of "highly scalable parallel" HMMER and BLAST that scaled effectively up to thousands of cores on a Cray XT5. By rearranging the input data sets of protein sequences, they were able to address I/O bottlenecking and other performance issues and ramp up their HSP-HMMER code to identify the functional domains of millions of proteins roughly 100 times faster than the currently available open-source MPI-HMMER.

In the spirit of teaching an almost 20 year-old dog new tricks, Jacqueline Addesa, a student from Virgina Tech, presented her argument for using the open source Haskell, a language usually reserved for big commercial applications, in an HPC setting. Addesa compared Haskell to C on multiple sequence alignment jobs, and found that the C code quickly became about 37 times longer than the same implementation using Haskell, with Haskell running about 2.68 times faster on hefty genetic sequence data sets.

Finally, a group from Brunel University teamed up with the folks at Petapath to demonstrate a power-efficient, high-performance heterogeneous computing system to simulate biochemical pathways, at the core of which is complex modeling using Ordinary Differential Equations (ODE). The team used a prototype accelerator card, Petapath's e740, to parallelize selected sections of the computational analysis to seriously reduce the job time of ODE tools like BioNessie, a biochemical network simulator developed at the University of Glasgow, and lower power consumption to roughly 3.84 GFLOPS of computation per watt.