Nathan_P wrote:OK, a simple question
The current donor pool for BA is small using the following hardware:- 4p is 4p G34, 4p LGA 2011, 4p socket F and the odd 4p LGA 1567. 2p is 2p G34, 2p LGA 2011, 2p 1366, 2p 1356 and the odd 2p C32.
The new core increase will remove 4p socket F, 2p LGA 1356, 2p LGA 1366, 2p C32, some 2p LGA 2011 and some 2p G34. How are less machines going to do more science?
You'll have to invent some numbers to fill in the following potential explanation.
Some percentage X of BA assignments are currently completed in A days by machines that will still be folding after the change(s) take place. The remainder of BA assignments are completed by slower machines in B days. The average completion time is X*A + (100%-X)*B. After the change, deadlines will be shorter and the average completion rate will be A rather than some weighted average of A and B. Yes, fewer WUs will be completed per day, but they'll be completed faster so they'll be more valuable scientifically.
By a similar argument, assuming some percentage of machines which will no longer be folding BA will be folding SMP (granted that that's an uncertain number) the average completion rate of SMP assignments will go up.
If my earlier suppositions are valid (still unconfirmed), the science done by both SMP and BA will improve, speed-wise.
Science dictates some minimum number of trajectories are valuable. Allocating more WUs than are needed is not good, but having trajectories that nobody is working on is also not good. The ideal situation is where science needs roughly the the same number of trajectories as there are machines working on those projects.
If science needs
more SMP trajectories to be completed than there are machines to work on them, some are not being worked on, slowing overall progress. If science needs
fewer trajectories than the number of donor machines attempting to fold them, either less-critical trajectories are added to the server or the server runs out of work (another cause for angry Donors).
If the FAH virtual supercomputer were pure hardware, no emotions would be attributed to it. Matching the active jobs to the hardware would be as easy as adjusting a few parameters, and nobody would object. Since individual Donors DO care about what happens to their points -- often very, very strongly -- tuning the virtual supercomputer gets very complicated. Strong emotions are felt by anyone who perceives a potential loss in their status. In situations like this one, unfortunately the PG has a goal of maximizing science which may not coincide with maximizing the status of every individual . Reconciling those two goals is NOT EASY. I wish them luck.