Einstein@Home and "Latent" Computing

At a strategic planning session on research support recently, I suggested we look into using our "latent computing" resources to work on research computing problems during off peak times. What I meant by this is that we have thousands of very fast processors which are sitting idle in our student labs between 11PM and 7:45AM every day. That's a chuck of nine hours for almost half of the machines in our labs. Originally, I was thinking that we should run Condor clustering on these after any updates are done to them each night. I was told that it's not clear that we want to reconfigure them each night and there are only a certain class of problems which run this way. While I agree with the latter point, I can think of a few massively parallel (MPP) or embarrassingly parallel (EPP) processing problems off the top of my head in which Penn State researchersare actively engaged (e.g. Monte Carlo simulations of radiative transport).

When I was at CalTech last week, one of those problems landed in my lap -- Einstein@Home. The great thing about Einstein@Home is that it works on a set of data which Penn State researchers have helped gather and are in the process of going through and cataloging. Einstein@Home operates on the LIGO Science Consortium's data runs looking for observations of gravitational waves. Rather than using something more tightly coupled like Condor, Einstein@Home uses BOINC -- The Berkeley Open Infrastructure for Network Computing. BOINC was developed for SETI@Home but has been generalized for problems which can run over networks of loosely coupled computers. Some of you have probably already used BOINC without knowing it when you loaded the SETI@Home screen saver or contributed to the Protein@Home effort. BOINC is lightweight and gives priority to everything else. What you don't use on your computer, BOINC can use.

We again may be confronted with the question, "But is it Green?" As I said in a previous post, if we design a system to lower peaks or even out energy utilization over the course of a day, while the system may use extra power, it is making use of power at times of traditionally low power consumption. Let's try to take advantage of our latent computing power with the goal of aiding Penn State researchers.

I would be interested in other BOINC-enabled, MPP, or EPP problems which you think would fit nicely into our down times, particularly if those problems would directly advance the cause of Penn State research. Please drop me a line or comment on this entry.