Message boards : Questions and problems : BOINC 6.10.58 released to public
Message board moderation
Author | Message |
---|---|
Send message Joined: 29 Aug 05 Posts: 15550 |
The new version is available from the BOINC download page. See the release notes for more information. Some of the changes for this release are: |
Send message Joined: 29 Aug 05 Posts: 15550 |
* New: Added ability for a project to be specified as a "Backup Project" by setting its resource share to 0 Projects require at least Changeset 20315. Projects supporting this: - AQUA - ClimatePrediction.Net (since June 4, 2010) - Collatz Conjecture - DNETC - Einstein@Home (since June 8, 2010) - Enigma@Home - Primegrid - Seti@Home - Seti beta If you know of more, let me know (through PM), then I'll add them here. |
Send message Joined: 28 May 10 Posts: 8 |
On my XP windows machine with an NVIDIA GPU, BOINC continually requests GPU WUs from projects that don't support a GPU (such as climateprediction.net and world community gird) and therefore gets no WUs. After a few days of doing this, it will finally download some non-GPU WUs. I upgraded from version 6.10.18 to the new release 6.10.56 hoping that would fix the problem, but it didn't. So ... what to do ... |
Send message Joined: 20 Dec 07 Posts: 1069 |
On my XP windows machine with an NVIDIA GPU, BOINC continually requests GPU WUs from projects that don't support a GPU (such as climateprediction.net and world community gird) That's not a bug, that's a feature (works as designed:-) and therefore gets no WUs. That must read: and therefore gets no GPU WUs. After a few days of doing this, it will finally download some non-GPU WUs. If it didn't so earlier, BOINC had its reasons (debt issues,...). I upgraded from version 6.10.18 to the new release 6.10.56 hoping that would fix the problem, but it didn't. So ... what to do ... Nothing, all is fine (in this respect). Gruß, Gundolf |
Send message Joined: 18 Jan 08 Posts: 36 |
* New: Added ability for a project to be specified as a "Backup Project" by setting its resource share to 0 It would be interesting to get a little description on the behavior provided for Backup Projects, covering such points as: 1. When the non backup projects have work, is any backup work prefetched? If so, how is it handled as it approaches deadline? 2. If not, then how far below desired must non backup work fall before any backup work is fetched? To zero work?, to first empty task slot?, to some percentage of desired? ... 3. If non backup projects work down to zero on local host, how much backup work is fetched? The full specified desired queue? some percentage of that? Only a task to fill a currently empty slot? something else? 4. Once desired work is again available, how is backup work in local inventory handled? For tasks already started? For tasks not started (if it does that)? I'm just generally interested, the specific questions are just meant to stimulate response--not necessarily the right questions to get a general description. Assuming the behavior turns out to be civilized, I like this addition, conceptually. Keeping everyone happy and believing it civilized is probably as nearly impossible as are other scheduler goals. Hence the questions. |
Send message Joined: 28 May 10 Posts: 8 |
[quote]On my XP windows machine with an NVIDIA GPU, BOINC continually requests GPU WUs from projects that don't support a GPU (such as climateprediction.net and world community gird) That's not a bug, that's a feature (works as designed:-) If a project does not support GPUs, then why waste time requesting GPU work??? and therefore gets no WUs. That must read: and therefore gets no GPU WUs. Well duh... |
Send message Joined: 29 Aug 05 Posts: 15550 |
1. When the non backup projects have work, is any backup work prefetched? If so, how is it handled as it approaches deadline? No. BOINC will only download work from that back-up project when either of the other projects do not have work or are temporarily out of the running. 2. If not, then how far below desired must non backup work fall before any backup work is fetched? To zero work?, to first empty task slot?, to some percentage of desired? ... When BOINC is out of work for your normal projects, or about to run out of work, it'll fetch work from the back-up project. 3. If non backup projects work down to zero on local host, how much backup work is fetched? The full specified desired queue? some percentage of that? Only a task to fill a currently empty slot? something else? I don't know. Am running that as a test with Seti down and Enigma as back-up. But Enigma still had work in queue, so I'll have to wait before I can answer that. Anyone else who knows? But I think it's only one task per CPU. 4. Once desired work is again available, how is backup work in local inventory handled? For tasks already started? For tasks not started (if it does that)? They'll just run every once in a while and before their deadline. |
Send message Joined: 20 Dec 07 Posts: 1069 |
If a project does not support GPUs, then why waste time requesting GPU work??? How should the client know if the project might have introduced a GPU application since the last request if it doesn't ask? Gruß, Gundolf |
Send message Joined: 29 Aug 05 Posts: 15550 |
If a project does not support GPUs, then why waste time requesting GPU work??? BOINC doesn't know which projects have GPU work and which don't. A project that has no GPU app today may have one tomorrow. If you want BOINC to use the GPU on that project, it'll fetch work automatically, when it next contacts that project and sees they finally have a GPU app. By the way, if you don't hammer the Update button, the back-off for the GPU scheduler polling projects will become longer and longer, eventually it'll only ask for GPU work from the projects once every 24 hours. |
Send message Joined: 29 Aug 05 Posts: 15550 |
3. If non backup projects work down to zero on local host, how much backup work is fetched? The full specified desired queue? some percentage of that? Only a task to fill a currently empty slot? something else? I thought wrong. :-) 29-May-10 03:39:33 | Enigma@Home | [sched_op_debug] CPU work request: 513681.29 seconds; 2.00 CPUs 29-May-10 03:39:36 | Enigma@Home | Scheduler request completed: got 37 new tasks Full 5 days worth of work. |
Send message Joined: 18 Jan 08 Posts: 36 |
Full 5 days worth of work. Grumble. I liked your guess better than I like the truth. Nevertheless it is good to know the truth. Thanks. |
Send message Joined: 5 Oct 06 Posts: 5123 |
Full 5 days worth of work. Is five days your CI or AW setting? For a backup project, shouldn't we ask for it to be CI only? In fact, I thought that was what David said when he checked in the change. I'll have to go back through the notes. Edit - [trac]changeset:20286[/trac]: no, he didn't. Memory playing tricks again. |
Send message Joined: 29 Aug 05 Posts: 15550 |
Edit - [trac]changeset:20286[/trac]: no, he didn't. Memory playing tricks again. Join the club. :-) CI = 0 AD = 5.1 Anyway, I sent it to BOINC Alpha as well. I'd expect 1 task per CPU from the back-up project, not the full amount of additional days... especially as they don't run anyway. |
Send message Joined: 30 May 10 Posts: 3 |
Dear Friends: I installed BOINC 6.10.56 on my Quad-CPU HP workstation with an NVIDIA GeFORCE 8800GTS GPU. OS is Windows 7-64 bits, NVIDIA driver version is the latest, 197.45. Unfortunately, when requesting new tasks for GPU, Milkyway@home replies: "Message from server: No work sent, Your computer has no ATI GPU, Your NVIDIA GPU lacks the needed compute capability (1.3, required for double precision math" I don't see why and where there is a problem: my GPU is listed as CUDA 3000-compatible. Please help. |
Send message Joined: 29 Aug 05 Posts: 15550 |
Not a BOINC problem, but a Milkyway requirement. They only allow GPUs which can do double precision floating point math to do their work. Your GPU apparently can't do this. I think only the GTX 2xx and 4xx versions can do double precision, as well as the Tesla. |
Send message Joined: 9 Jun 10 Posts: 2 |
I keep getting warnings about "Suspending computation - CPU usage too high" and "Resuming computation", which appeared only since version 6.10.54. I tought it was part of the problems with .54, but they reappeared when I installed .56 Is that normal, due to new features? |
Send message Joined: 29 Aug 05 Posts: 15550 |
See Suspending computation - CPU usage too high for more information. |
Send message Joined: 9 Jun 10 Posts: 2 |
Perfect, thanks... |
Send message Joined: 18 Jan 08 Posts: 36 |
See Suspending computation - CPU usage too high for more information.One extra detail I believe I've seen, which is actually quite annoying. As Einstein (the second most popular BOINC project, I believe) does not yet use web site code which gives you the option to set this throttling function as you like, or disable it, you have to do so from SETI or another project which does. This point is covered in the link Ageless provided. The nasty additional surprise for me was that if I subsequently changed a preference using the Einstein web page (in my case I raised a parameter governing my work queue size), then when my host updated preferences from the Einstein site (as it must, since the freshness date is now newer), the undesired (to me) 25% returned. In other words, setting 0 for this parameter at SETI does not currently propagate to Einstein (and most likely any other project running too old a version). So, despite the fact I've embraced the new "backup project" capability to lower SETI from 4% to backup, I must do any and all general preferences adjustments at SETI, and not at Einstein, as I wish not to lose productivity and wish not to have a message log full of start/stop messages from this source. Fortunately when times are good I seldom change general preferences. |
Send message Joined: 29 Aug 05 Posts: 15550 |
As Einstein (the second most popular BOINC project, I believe) does not yet use web site code which gives you the option to set this throttling function as you like, or disable it, you have to do so from SETI or another project which does. Einstein does use the updated web code. It has done so since the 8th of June. See Einstein global preferences. See also Web code update and the news on their front page: Web code update ...then when my host updated preferences from the Einstein site (as it must, since the freshness date is now newer), the undesired (to me) 25% returned. In other words, setting 0 for this parameter at SETI does not currently propagate to Einstein (and most likely any other project running too old a version). It's the other way around. Your preferences only propagate to other projects as and when your BOINC makes contact with that project's scheduler. It doesn't do it on its own. So if you change something at the Einstein project and then do an update, before BOINC has had a chance to report work for instance, it'll download the new preferences from Einstein, since those are the ones you changed last! |
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.