stuck in loops
12 Oct 2010 3:06:57 UTC
Topic 195376
(moderation:
Hi,
While it is interesting to participate the Einstein@home project, I came across a couple tasks that were stuck in loops. I could see them (CUDA) using up hours and hours of GPU calculation, but progress remained at the same point. Hope you guys can resolve or prevent this type of problem soon. Thanks.
Henry Tsai, Ph.D.
stuck in loops
)
If you see one of these units in progress, just exit Boinc and then restart it, it should reset the numbers to actual time used and pick up crunching from where it stopped. Apparently there are no clear cut answers as to why this happens but one idea is that it could be related to the Boinc benchmarks being run every week. The only problem with that is that I just had one crash, on another project, and the benchmarks were not run. Apparently Boinc does not stop the gpu when the benchmarks are being run, causing issues.
RE: Apparently Boinc does
)
Where did you read that? I haven't seen it reported anywhere, and it doesn't match my observations: I always see NVidia tasks being suspended when benchmarks run, although I can't check that the same is true for ATI (no suitable card here).
One recent change, which David actually checked in at my request, is that GPU tasks are no longer unloaded from the graphics card's memory during benchmarking (it's only the CPU which is benchmarked, after all). In theory, that should mean that the GPU task can restart more quickly afterwards - but maybe it causes problems with the hybrid Einstein app. If that's the case, it should be confirmed and reported on boinc_alpha.
my computer is also stuck in
)
my computer is also stuck in a loop, the project is not advancing. turning my computer off and on does not help
RE: RE: Apparently Boinc
)
Read this http://boinc.thesonntags.com/collatz/forum_thread.php?id=603
Slicker is the Project Admin over there, scroll down a few messages to number Message 9881 - Posted 4 Oct 2010 13:37:13 UTC.
Is that in version 2.05? Also the Boinc Alpha mailing list is not always a nice place for us 'boinc users' especially if we seem to be whining! Being ignored on the list seems to be a fairly common occurrence. Of course impressions are what is important to us non programmers, not always deeds, so if one gets the impression they are being ignored, even if they are not, it tends to stick. The mailing list has never been what is called 'user friendly'. And NO I am not blaming anyone, it is what it is.
RE: my computer is also
)
Just exit Boinc and then after a minute or so restart it and it should be okay. If not just abort the task and move on, the task will be sent to someone else. Not all tasks can be crunched, some are just bad.
RE: RE: RE: Apparently
)
I was thinking of BOINC changes. This was, to be specific, changeset [trac]changeset:20765[/trac] - actually, at the beginning of March, so not that recent. That's been in every BOINC client since v6.10.43, so I think we'd have heard of problems by now if they were widespread.
v2.05 is a Collatz-specific application version - it sounds as if they have a little local problem there, different from any generic BOINC problem. Doesn't stop Einstein having a local version of the same problem, of course - and sometimes these "two projects at once" reports do reveal underlying BOINC problems.
I hear what you say, and agree with the sentiments. But for the time being, boinc_alpha is what we've got, and if we want to improve BOINC, we can only work with it.
RE: RE: RE: RE: Appar
)
I did not know 2.05 was a Collatz thing, sorry about that. But the gpu problems are occurring across several projects, not just Collatz. I have also personally experienced it on DNETC too. The current theory is that when the benchmarks are started by Boinc it is causing a freezing of the gpu and the workunit is not restarting itself afterwards. Although my personal experiences do not seem to reflect this, I attached to DNETC with an Nvidia card and the same evening it froze! The next day when I again checked the machine, I have a bunch, I noticed that the unit was over 20 hours in and that is VERY long for those units, so I stopped and restarted Boinc and the unit went back to only 15 minutes, roughly, of crunching time and then errored out after a few more minutes.
I know Dr A and crew are what they are, I am not slamming them anymore, I am just relating my own personal experiences. I encourage others to make their own, who knows they could be much different. And to be honest Rom and others can be OUTSTANDING!!