Bigger isn’t better
01.10.12 - Permalink
For some complex biological structures, bigger isn't always better. Instead what most researchers want to explore is the actions of smaller biological structures over longer periods of time.
Researchers from the University of Utah, University of California at San Diego, Rutgers University, Stony Brook University, and the University of Florida have teamed up to get a closerand longer lookat those smaller proteins through the National Science Foundation's Petascale Computing Resource Allocations (PRAC) program. The PRAC program support science teams as they work with NCSA to run on Blue Waters and other petascale systems.
Molecular dynamics research captures the interactions of individual atoms in complex biological structures like proteins, DNA, and RNA. Looking at smaller systems of atoms isn't necessarily easier than looking at ever-larger systems, explain the researchers, because molecular dynamics is very difficult to parallelize as you can't parallelize for time.
A researcher wrings performance out of today's supercomputers by breaking a problem up into many small pieces and having those pieces run simultaneously on the computer's many processors. Because the atoms in the simulation are influenced by what has happened at previous points in the simulation, you can't have the future proteins or RNA being simulated at the same time as their earlier selves. You have to take it one timestep at a time.
So the team is looking at a process where you have lots of simulations, but the processors share moderate amounts of information that are really crucial to solving the overall problem. Using this intermediary approach, the team runs multiple copies of the same system under different conditions, simulating the system at a variety of high temperatures, pH levels, or even physics models not typically found in nature, for example. These copies swap data every so many timesteps within the calculation.
People have been using this methodcalled replica exchangefor about a quarter of a century. However, with recent access to large-scale HPC resources through the TeraGrid and XSEDE, the use of these methods has exploded. In fact, the team and others have seen a series of successes by using it. On Blue Waters and other systems of its size, the team wants to begin enabling multidimensional replica and information exchange between ensembles of independent molecular dynamics simulations, delving into the effects of changes to more than one factor that influences the system's behavior. They've learned that only looking at one factor isn't enough.
As part of the PRAC program the team is currently updating the simulation code frequently used to do that work. Called AMBER, more than 1,000 sites around the world take advantage of the code.