I recently had a rouge process fully utilizing one of the 8 threads on the box with my GTX560 in it.
This would be annoying enough in any event; but for some reason despite the CPU processes having a lower priority than the GPU process the latter was consistently starved of CPU time to the point that instead of 35-40 minutes per BRP4 CUDA task it was taking as many as 8 hours to complete one.
Other than keeping a closer eye on things and finding a replacement for the app if its newest version goes crazy as well is there anything simple I could do to force better behavior when the scheduler is overloaded? I'd prefer not to have to much with an app_info file again; the headaches involved with each app upgrade are really annoying.
Copyright © 2024 Einstein@Home. All rights reserved.
Windows scheduler headache
)
Use Linux instead... ;-)
One trick would be to run a timed job to look at the current elapsed times and email you or pop up a dialog to warn you of slow progress...
Then again, that could be a feature request for the Boinc manager to note unusually slow progress...
Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)