Had to limit # concurrent tasks due to "hi priority" WCG downloads

Message boards : Questions and problems : Had to limit # concurrent tasks due to "hi priority" WCG downloads
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Joseph Stateson
Volunteer tester
Avatar

Send message
Joined: 27 Jun 08
Posts: 641
United States
Message 91981 - Posted: 29 Jun 2019, 18:39:48 UTC
Last modified: 29 Jun 2019, 18:40:35 UTC

Fixed, but do have a question.

First want to show the problem.
One of my systems had been doing only GPUGRID and I allowed WCG tasks as I solved the CPU cooling problem. Unaccountably they were all deadline'd for 48 hours even though these were fresh downloads. With 12 threads that was not a problem as they would all complete but I did notice that my other BIONC system did not have this problem which was caused by all of the WUs having about 48 hours to finish on this system. I guess this was my luck and I got the "last" of the bunch.

One of the two GPUGRID work units (have pair of gtx1070) was "ready to run" so I decreased the # of cpus to allow it to run. That did not work. Only when I stopped all of the WCG tasks did the second GPU task start up.

Seems the high priority tasks preempt other tasks but there is no consistency. With all resources set at %100 resources either both GPU tasks should run or none should run if hi priority really preempt even with unused CPU threads. Why was just one allowed to run? In addition I had SETI gpu tasks in the queue but none of them got that left over GTX1070 either.

Some thought: Be nice if BM had a form to set various configurations thus avoid editing app_config. Alternately, a separate program to set parameters on various projects. I looked through some of the sources dealing with RPC calls but did not see anything related to setting a value in app_config but I might have missed it. I used project_max_concurrenet to ensure that only 8 tasks run and now my pair of GPUGRID are crunching in addition to the 8 WCG.
ID: 91981 · Report as offensive
Profile Joseph Stateson
Volunteer tester
Avatar

Send message
Joined: 27 Jun 08
Posts: 641
United States
Message 91986 - Posted: 30 Jun 2019, 3:48:54 UTC - in response to Message 91981.  

Unaccountably they were all deadline'd for 48 hours even though these were fresh downloads


I am guessing these were "lost tasks" that were resent as I had problem with this system and replaced the power supply Alternately, they were on the disk drive all this time but didn't show up as the WCG account file was corrupted. This system had corrupted gpugrid account & statistics files and the work units did not show up. However, I don't recall seeing any WCG error messages and I saw plenty of error message from gpugrid.
ID: 91986 · Report as offensive
mmonnin

Send message
Joined: 1 Jul 16
Posts: 146
United States
Message 92125 - Posted: 10 Jul 2019, 11:57:05 UTC

Use an app_config to set cpu usage on the GPU tasks to 0.1 CPUs so they will still run even if CPU tasks become high priority. This does not change how much CPU the GPU tasks will actually use. You can reduce the # of CPU tasks via BOINC Mgr setting or Process Lasso to allow GPUGrid tasks to get all the CPU cycles it needs.
ID: 92125 · Report as offensive

Message boards : Questions and problems : Had to limit # concurrent tasks due to "hi priority" WCG downloads

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.