- Joined
- Jan 9, 2006
This reminds me of a problem I have always had with FAH.
If our clients have a built-in cache, then why is it never used? If my GPU returns work units in 3 hours, and the deadline for all of them is 3 days, than why not let me download 10 of them, so that if my connection DOES go down, then I can keep folding for a day or two at least.
Years ago, it was that way.
Two problem were found:
1) Lots of WU's were not returned - just lost.
2) Some folders and teams cached their work units, so they could dump them all at once. That's counter productive for Stanford. They need that steady stream, and quick turnaround, to be efficient.
At that point, Stanford made a decision to just keep everything on-line for the folders, (more than they had in the past), and have just one unit at a time go out, and come back. Happily, this coincided with a big expansion of DSL, cable, etc, ISP services, where the user's were always on-line.
If you compare FAH's servers up-time, (as a group), with say, SETI, you can appreciate what FAH has been doing behind the scenes, to keep the WU's flowing.