Message boards :
Number crunching :
V7 & cuda
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
+1 Cheers. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Switched over to the new app_info this morning and received 64 new cuda50 tasks as expected with no v7 CPU units. Now if only a way can be found to pre-split AP tasks at their origin so that when they get to Berkeley they are on their own tapes and we can have a better supply of them. It seems to be the BOINC client combined with Murphy's Law. One of the last things an app does before exiting is write an empty "finish file", and that error indicates the app didn't exit within ten seconds after writing that file. I think the error was likely caused by shutting BOINC down after the file was written but before the app had gotten to its usual exit, although there's only a very brief time period between. IOW, you'd probably never see it happen again even if you tried to duplicate the conditions. The stderr for that task does show that BOINC actually restarted it after it had already finished, so I'm unsure of that analysis, but IMO that also indicates the problem is in the BOINC client. Joe |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Switched over to the new app_info this morning and received 64 new cuda50 tasks as expected with no v7 CPU units. Now if only a way can be found to pre-split AP tasks at their origin so that when they get to Berkeley they are on their own tapes and we can have a better supply of them. I've delved quite deeply into this kindof strange symptom over time, of which there are a large number of different odd behaviours originating from a similar set of root causes. It's taking me longer than I hoped to document those causes for (hopefully) wider benefit, but the general gist is some assumptions made about process and IO management that may have applied in the distant past, no longer hold true (if they ever did) under operating systems & runtimes optimising for desktop performance. Where low priorities and high system contention mix, such as might be encountered at system shutdown, IO completion, normal process termination & garbage collection can be postponed from the more usual fractions of a second, to on the order of minutes, on otherwise healthy systems. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.