The new credit system isn't using the rsc_fpops_est / rsc_fpops_bound specified when the workunit is created anymore. Is there new estimated time field for workunits?
On May 8, 2011, at 3:42 AM, robert miles wrote: > Then would simply setting the estimated time to 4 times as much when you > build the workunit provide enough wiggle room, or would you need somewhat > more than that? > > From: Travis Desell > > > It looks like the estimates for the GPUs are killing things about 1/3rd of > the way into the workunit. They have a time limit, it's just not high > enough: > > 07/05/2011 22:43:54 Milkyway@home Starting > de_separation_10_3s_fix10_1_287737_1304804381_0 > 07/05/2011 22:43:54 Milkyway@home Starting task > de_separation_10_3s_fix10_1_287737_1304804381_0 using milkyway version 62 > 07/05/2011 22:45:25 Milkyway@home Aborting task > de_separation_10_3s_fix10_1_287737_1304804381_0: exceeded elapsed time limit > 89.484881 > 07/05/2011 22:45:26 Milkyway@home Computation for task > de_separation_10_3s_fix10_1_287737_1304804381_0 finished > > Maybe clients should be updated to have a bit more wiggle-room when it comes > to early termination of a workunit? > > > On May 8, 2011, at 2:07 AM, robert miles wrote: > >> From: Travis Desell <[email protected]> >> >>> Getting rid of the entries with negative variance seemed to fix the WUs >> getting >>> immediately timed out (at least on DNA@Home), but on milkyway@home it >>> looks like the problem is still happening on the GPU applications.... >> >> Are the milkyway@home GPU applications the type where the CPU part first >> stays busy a while loading the GPU part into graphics memory, then goes >> into an almost idle mode for most of the rest of the workunit where it does >> little except checking whether the GPU part says it's finished, then finally >> goes >> into a mode where stats busy unloading the outputs from graphics memory? >> >> If so, I'd expect some large changes in any speed calculations that don't >> restrict the CPU operations observed to just one of those modes. >> _______________________________________________ >> 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. > > > ---------------------------------------------------------------------------------------------------------- > Travis Desell <deselt @ cs.rpi.edu> > 1-518-867-1054 > Adjunct Professor & Postdoctoral Research Assistant > Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA > http://www.cs.rpi.edu/~deselt/ > MilkyWay@Home ( http://milkyway.cs.rpi.edu/ ) > DNA@Home ( http://dnahome.cs.rpi.edu/ ) > Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ ) > ---------------------------------------------------------------------------------------------------------- > > ---------------------------------------------------------------------------------------------------------- Travis Desell <deselt @ cs.rpi.edu> 1-518-867-1054 Adjunct Professor & Postdoctoral Research Assistant Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA http://www.cs.rpi.edu/~deselt/ MilkyWay@Home ( http://milkyway.cs.rpi.edu/ ) DNA@Home ( http://dnahome.cs.rpi.edu/ ) Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ ) ---------------------------------------------------------------------------------------------------------- _______________________________________________ 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.
