I'm happy to announce that we will be resuming Human Proteome Folding 2 (HPF2) processing most likely some time this weekend. We have tested the new Rosetta code behind the scenes and the initial results have been validated by the researchers. Then next step is to launch it on the full grid to get a large enough set of results for the researchers to perform the final full-scale result validation. We think this will go well and we will be able to just continue running. The UD agent version happens to be ready first, so we will launch that shortly. The Boinc agent version is almost ready and just needs a little bit more work and testing. So we may launch it within a couple of weeks or perhaps earlier, depending on how everything goes behind the scenes. By the way, the results returned to date, from before the pause in the project, are valid and will be used by the researchers. Thank you to all of the members for your valuable contributions to this and the other projects!
Some of our more technically oriented members probably know about the result checking process we use. Essentially, each work unit is normally sent to more than one machine and we look for majority equality in the results before sending them on to the researchers. HPF2 is one of our projects that can make use of a statistical clustering scheme to find valid results from a large number of protein structure predictions. This method also permits the researchers to discard errant results. Therefore, for HPF2, we can have each member compute a unique result and this can accelerate the overall project by a factor of 3 or more. In other words, we don't need to run duplicate computations and thus can run more work through the grid in a given amount of time. The ultimate result validation burden remains with the researchers, as before. But we won't have to run as many work units through the grid. However, there will still be some preliminary checking of the results at our servers for UD and Boinc, to make sure the computations have not been obviously corrupted due to hardware errors or other problems. Members running Boinc will see that this scheme lets Boinc "validate" results sooner, thus awarding points sooner, while keeping the points assignment fair using a run duration normalization technique. If this approach works out well, we will implement it on other projects for which this scheme can be used. It will help take years off the time required to run massive projects such as HPF2.