Folding@home client can now run on supercomputers

Earlier this week at the supercomputing SC11 conference, researchers from the long-running Folding@home project demonstrated a new distributed framework named Copernicus that’s designed to allow F@h to scale across geographically-dispersed supercomputing clusters. It’s a logical step for a project that, over time, has grown to support GPUs, PS3s, and all major operating systems — but leveraging the power of an HPC cluster requires more than an OS port and a few design tweaks.

That’s where Copernicus comes in. F@h’s primary mission is to understand each step in the protein-folding process. Copernicus, in contrast, lets the user specify the desired end results, rather than a detailed, step-by-step description of how to obtain them. The Copernicus run-time, taking into account all of the available computational resources, then breaks down the desired end results into specific, efficient task packages.

Based on this diagram, my earbuds are actually proteins

As an analogy, consider balloon animals (without any creepy clown references). Folding@home focuses on examining every fold and twist that transforms a balloon into a giraffe, a rabbit, or Optimus Prime. It also measures the mistakes that occur when a balloon doesn’t fold properly, and what the resultant mutant creature looks like.

Copernicus allows researchers to start by describing the characteristics of the balloon animal they’re looking for. The software is then designed to scale across supercomputing clusters and leverage their massive parallel processing capability to locate best-fit balloons. One of the primary challenges, according to Vijay Pande’s team, was ensuring that the software could scale effectively across wide deployments.

“It opens the door to huge crowds of people using these methods, which have matured with Folding@home,” Pande said. “This method should be able to use any supercomputer on the planet completely,” Pande said. “Strong scaling to these extremes is unusual.”

In a discussion thread in the Folding@Home forums, Pande stated that, in essence, Copernicus is Folding@home for supercomputers — but went on to note that the project’s primary goal was to “take all of the scientific methodological advances in FAH’s backend and bring it to other researchers.”

The researchers behind FAH likely wouldn’t argue with more processing power, but as of today, Folding@home has an actual output of some 9.075 x86 petaflops. The combined output of the network’s seven million-plus CPUs would put it just under the current TOP500 supercomputer, RIKEN’s K.

Folding@home client can now run on supercomputers