Joined: 7 Jun 07
It is obvious from the continuous flow of \"out of work\" and \"no work for platform\" complaints, that the current way of distributing work is a constant source of irritation for many participants. Now that more personnel will become available, I think the project management should reconsider the usage of the \"homogeneous reduncancy\" mechanism, which afaik is at the basis of this problem. It can be tweaked a little, but it is becoming more and more obvious that this will never lead to a stable balance between machines of all platforms. I see these alternatives, maybe there are others:
- Use a quorum of 1 (or in other words, let a work unit be handled by only 1 participant). In that case some sort of sanity check on returned results may have to be implemented.
- Use a floating point library that ensures that the application returns the same result no matter what platform it runs on. Drawback is a possible loss of performance, which might be compensated by increased effort by happy participants :D
BOINC.BE: For Belgians who love the smell of glowing red cpu's in the morning
Joined: 11 Dec 07
It is obvious from the continuous flow of \"out of work\" and \"no work for platform\" complaints, that the current way of distributing work is a constant source of irritation for many participants.
I was just about ready to post a message about how the tweaks to the feeder code that were done appear to have not helped a bit. I think there was also a message about possible new HR classes. If that has been implemented, it also didn\'t help.
IMO, the project needs to get in touch with someone with BOINC and who has HR experience. Perhaps they have done so, but the silence is deafening...