Sorry for Top Posting... When this silly Cuda stuff was started the assumption was that Cuda would fit inside the current scheduler so that if 4 CPU's were identified it would "fit" inside the 4 CPU Scheme. In Seti there is a silly user that has 4 - 295's that complains that the scheduler will not give work for the 4 295's and is limited to the 4 cores that is identified by the BOINC Client.
So to say that while developing a New Plan class on a flawed system is okay and not worth the time or resources. Someone need to rethink for a moment. Both the Scheduler and the Client do not properly deal with the existing Cuda instance. Then Multithreading is already underway in various projects. Various issues have been identified here, what is the scheduler supposed to do for those instances. A Blade Server with 48 CPU's or one of Nvidia's Tesla S1070. Something has to get a bit smarter. We are no longer dealing with a single computer. At 03:18 AM 9/11/2009 +0400, Raistmer wrote: >Well, I just got evidence that this work in 6.10.4 indeed. >It makes my proposal much less actual right now of course , cause there is >only hypothetical "unsupported accelerator", but maybe it's worth to >implement it anyway for future hardware... > > ----- Original Message ----- > From: David Anderson > To: Raistmer > Cc: BOINC Developers Mailing List > Sent: Friday, September 11, 2009 3:01 AM > Subject: Re: [boinc_alpha] BOINC 6.10.4 abuses high priority mode > > > Raistmer wrote: > > >If your app_info.xml specifies that app versions > > >use coprocessors that the client knows about (NVIDIA or ATI) > > >then jobs using those app versions will always run, > > >regardless of CPU usage. > > Well, unfortunately, AFAIK , job for those app versions will never run > > if project not support same coprocessors by separate work feed. Maybe > > something changes in that aspect in last BOINC versions and I just don't > > aware about it? > > If app_info.xml says that an app version uses CUDA or ATI, > then the client will schedule jobs accordingly > even if the project doesn't have CUDA or ATI apps. > If it doesn't, it's a bug; please send me message log > with <cpu_sched_debug> set. > > > I mean if project not use let say ATI coprocessor for its stock app then > > it will not give separate work for that coprocessor then app_info > > containing reference to this coprocessor will always recive "No work" > > message from project servers, correct? > > No; it should get work if the project has any. > However, this was fixed fairly recently ([18994], 2 Sept) > and I'm not sure that's in 6.10.4. > > -- David > >_______________________________________________ >boinc_dev mailing list >[email protected] >http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev >To unsubscribe, visit the above URL and >(near bottom of page) enter your email address. _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
