How Important are Deadlines?
From FaHWiki
You need a very long series of computations to simulate the folding of a single protein, These calculations are fundamentally serial, not parallel. (You can't simulate a particular time until you know what happened at a previous time.)
Most other distributed computing projects are fundamentally parallel. e.g.- for Seti@Home, you can analyze one part of the sky while I am analyzing another part of the sky. As long as all points get analyzed, there is no need for them to be completed in any particular order.
FAH has managed to parallelize portions of the simulation so that a number of computers can all be working on the same protein at the same time, but there still are very significant parts of the analysis where the next WU cannot be generated until you finish the one you're working on. There are a limited number of WUs for a given protein. When you're holding a WU but not working on it, the next person is waiting for you to finish.
If everybody takes a WU and only works on it half of the time, then the total simulation will take twice as long to complete. If that process takes (say) 6 months, and because people do not return the results quickly, the process isn't completed for a year or more, the researcher may have moved on to another project or graduated or whatever.
Just because your hardware is fast enough to finish a WU in a very small fraction of the deadline, you shouldn't delay returning the result until near the deadline. The deadlines are there just to take care of WUs that get lost, not to give you an indication of when the project needs the results.
On the Folding-Community Forum kasson of Pande Group posted about the work unit generation:
After the work units come back, the results are stored on our servers and we use them to analyze the protein-folding events we are simulating. One of the reasons it is helpful to get work units back quickly is that many problems we study occur over long timescales, so we have to chain together many sequential work units (as well as perform a number of statistical analysis tricks). As a gross measure of scientific progress, a link to papers we have published using Folding@Home results is here: http://folding.stanford.edu/papers.html
Related Links
