I've built a couple of clusters with ROCKS. One that has 156 cores and one test bed with 20 cores.

The reason why f@h doesn't work very well on the ROCKS roll is because of how the f@h client is built. F@H does some cloud computing algorithms that basically turn each box running the client into a worker node. When you finish your work order, you sent the results back to the server farm that stores the information (head node for the cloud).

When you set up a cluster, it's the same principle -- just smaller scale. Now you can poke around in the scheduling program in the ROCKS roll (either condor or fork -- fork would work better for you since it's a one user cluster) you can have it take each WO that you get from Standford and split it up again (might have to ask the boys over there for some help) so that each worker node gets a piece of the pie, then you would be able to efficiently use a cluster for folding.

**I'm writing this while working on some DNS issues with our cluster...so I'll clean it up and fix any errors when I get done**


James