Page 3 of 3
Re: Selective Folding (branched from Nvidia issues forum)
Posted: Tue Jun 09, 2009 2:29 am
by v00d00
As far as im aware the networking issue was solved some time back.
Nowadays if you try and forceably reroute a connection to a preferred server the client just hangs. You are assigned a workunit at the AS, then you go pick it up from whatever server its available on. If you try and reroute to a different server the client assumes you have a network issue and hangs until it can retry.
Re: Selective Folding (branched from Nvidia issues forum)
Posted: Wed Jun 10, 2009 5:16 am
by shdbcamping
Amaruk and Bill 1024 both pointed out that in one post I said I dumped WU and in another post I said that I don't. To clarify: After reading the feedback on why it was a bad idea to dump WU, I stopped discarding WU and am now shutting down half of a 9800GX2 *temporarily* as necessary to control temperatures. Consider it a manual form of throttling back the GPU. I admit that I still lose quite a few WU this way when they go over deadline, but I do my best to fold everything they send me, which I would hope that we could all agree is the best thing for the science.
I am trying to conform to the Project's "Everyone must run everything policy", which I do by accepting the AS WU's. I simply must also protect my equipment as well, and will continue to do so until the underlying cause of the overdue WU's can be remedied. Unfortunately for the Project it also delays the science as well.
The rest of the explanation for this is in the other thread that I started specifically to try to get a resolution to GX2 Heat issues Here: viewtopic.php?f=52&t=10128&start=105#p101639 .
Re: Selective Folding (branched from Nvidia issues forum)
Posted: Sun Jun 14, 2009 5:14 pm
by fractal
I contribute to several CUDA GPU folding projects. Most of my GPU's are on FAH, but I do dabble on any DC project that comes out. One of the others has a similar problem to what is described here. They too want the units returned as soon as possible, but the framework (BOINC) gives 4 work units to a 4 core machine even if it only has one GPU. The project gives a bonus to work returned within one day. So I, like many, abort the 3 extra's when I see them and am close to 90% finished on the current one. The good news is that the project IMMEDIATELY sees they were aborted and reissues them. The bad news is I sat on them for half a day with no intention of crunching them. But, aborting them after a half a day is better than sitting on them for three days while working through the queue. This is a flaw in the framework.
On FAH, the issue of cherrypicking has been around for donkey ages. Doners can not say "don't give that to me", Donors can not say "take it back". They can only be bad or worse. Evil donor, evil. Yeah, right.
Don't get me wrong, It is not worth my effort to cherry pick wu's. I fold whatever I get. But, competitive donors are NOT going to go away, so you had better figure out a way to let them indicate they are not going to process a work unit without harming the science. Otherwise, don't blame THEM when the science is hurt by your broken framework. Don't blame people who's hardware won't tolerate some types of work. Don't blame donors who's conscience won't allow them to work on a certain project. Donors should be able to pick the projects they want to work FOR WHATEVER REASON without being accused of "harming the science".