Forums :
Technical Support :
No CPU time again
Message board moderation
Author | Message |
---|---|
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
Not sure why I'm bothering to ask in here as I never get an answer, but one of my hosts is idling about. http://www.cosmologyathome.org/results.php?hostid=452344 I thought I'd sorted the nonsense with VB tasks, by downgrading VB to 5 instead of 6, but one host is taking 14.5 minutes to do bugger all CPU time then sending them back, half of those are marked as ok on this end for some reason and half as a computation error. |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
I'll reply to myself since nobody's in here. Don't let this run on lots of cores. I have three 24 core machines. 24 cores on this project doesn't work. You need to limit it to say 6 cores and let it do 4 of them, using app_config in the Cosmology project folder: <app_config> <app_version> <app_name>camb_boinc2docker</app_name> <plan_class>vbox64_mt</plan_class> <avg_ncpus>6</avg_ncpus> <cmdline>--nthreads 6</cmdline> </app_version> </app_config> |
Nflight Send message Joined: 4 Aug 07 Posts: 7 Credit: 1,691,791 RAC: 0 |
Peter your posts are appreciated, thank you for the knowledge you share, which can make a difference to those who don't have all the answers, you are a God Send.!! ![]() |
poppinfresh99 Send message Joined: 1 Mar 22 Posts: 21 Credit: 1,227,043 RAC: 1,032 ![]() |
Instead, you could probably change “Max # CPUs” on this project’s settings webpage. |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
Instead, you could probably change “Max # CPUs” on this project’s settings webpage.I wasn't sure if that was CPU cores per task or CPU cores per computer. I want to use all the cores, just run each task on 6 of them, otherwise it fails. |
poppinfresh99 Send message Joined: 1 Mar 22 Posts: 21 Credit: 1,227,043 RAC: 1,032 ![]() |
I think “Max # CPUs” is for the whole project. I also thought that <avg_ncpus> was for the whole project... https://www.cosmologyathome.org/faq.php#limit-cpu https://boinc.berkeley.edu/wiki/Client_configuration#Project-level_configuration That is, for the whole application (not project), but camb_boinc2docker should be the only application we are running. |
.clair. Send message Joined: 4 Nov 07 Posts: 640 Credit: 13,333,595 RAC: 21,152 ![]() |
Max # jobs - is for project = Limit on "Tasks in progress" possibly per computer Max # CPUs - is for each work unit Each has a limit of 8 on the pref`s page It was discovered not long after cam2docker started that VB goes mental if more than 8 cpu`s are used and many tasks have errors for no good reason , unless you are lucky . Max # jobs - 8 { 1<8 or "No Limit" = 300 ish } I did`nt count them Max # CPUs - 6 {1<8 - don't use "No Limit" } should have a 24 core system - 4 "running" work units and - 4 "ready to start" And no need for an "app config" file . 7 cpu`s takes 10<11 minit on my opteron16 |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
7 cpu`s takes 10<11 minit on my opteron16Pah! I'm doing a job in 3 minutes on an i5-8600K, and 4 at once in 6 minutes on a Ryzen 9 3900XT. I SHALL PREVAIL!!!! |
poppinfresh99 Send message Joined: 1 Mar 22 Posts: 21 Credit: 1,227,043 RAC: 1,032 ![]() |
Max # jobs - is for project = Limit on "Tasks in progress" possibly per computer This is great to know! I tested it, and I was wrong about "Max # CPUs". It was discovered not long after cam2docker started that VB goes mental if more than 8 cpu`s are used and many tasks have errors for no good reason , unless you are lucky . It's great to know that "Max # CPUs" is needed to fix this and fixes this so nicely! Is this more-than-8-CPU bug true for VB in general (that is, not just for Cosmology@home)? Maybe it's just for CERTAIN VirtualBox situations?? https://forums.virtualbox.org/viewtopic.php?f=2&t=87445&start=15 |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
Is this more-than-8-CPU bug true for VB in general (that is, not just for Cosmology@home)? Maybe it's just for CERTAIN VirtualBox situations??I don't have any problems running loads of VB except: If it's the computer I'm trying to use for web browsing etc, more than half the cores on VB grinds it to a halt - it's fine on a Boinc only machine though. Cosmology needs to run on 6 cores for each task, not 24 each. But I can happily run four of 6 core on a 24 core machine I'm not using for anything else. Virtually no errors. |
.clair. Send message Joined: 4 Nov 07 Posts: 640 Credit: 13,333,595 RAC: 21,152 ![]() |
7 cpu`s takes 10<11 minit on my opteron16Pah! I'm doing a job in 3 minutes on an i5-8600K, and 4 at once in 6 minutes on a Ryzen 9 3900XT. I SHALL PREVAIL!!!! Yup , the opteron iz little faster than an Athlon XP3200 in single threaded crunching [benchmarks] , I know this coz mine only died last year crunching Einstein {I don't like chucking stuff if it still workz} Hopteron is / waz a cheap webserver cpu in its day , but crap at math . It waz a cheep system {ebay parts} when I got it , I soon found out why . |
.clair. Send message Joined: 4 Nov 07 Posts: 640 Credit: 13,333,595 RAC: 21,152 ![]() |
Max # jobs - is for project = Limit on "Tasks in progress" possibly per computer From reading the posts on VB and the linked bug trak , that is the kind of stuff that waz happening here on cosmo when CB2D started , so several upto 8 core tasks ran a lot faster than one `big` one , if it ran at all . |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
I like to keep old stuff running too, even though it may be more sensible to chuck it and buy something that uses less electricity.7 cpu`s takes 10<11 minit on my opteron16Pah! I'm doing a job in 3 minutes on an i5-8600K, and 4 at once in 6 minutes on a Ryzen 9 3900XT. I SHALL PREVAIL!!!! I've never managed to break a CPU though, MBs occasionally expire, and GPUs expire a lot. |
poppinfresh99 Send message Joined: 1 Mar 22 Posts: 21 Credit: 1,227,043 RAC: 1,032 ![]() |
I also thought that <avg_ncpus> was for the whole project... I now see that the FAQ linked to (in above quote) also has <max_concurrent>1</max_concurrent> in addition to avg_ncpus, so, what Peter said about just using avg_ncpus (without setting max_concurrent to 1) should work. Though “Max # CPUs” on this project’s settings webpage is still easiest :) *If* someone also wanted to set max_concurrent, I think "Max # jobs" on this project’s settings webpage would work and be easiest. |
poppinfresh99 Send message Joined: 1 Mar 22 Posts: 21 Credit: 1,227,043 RAC: 1,032 ![]() |
*If* someone also wanted to set max_concurrent, I think "Max # jobs" on this project’s settings webpage would work and be easiest. Though I just learned that "Max # jobs" and max_concurrent are a bit different... - "Max # jobs" is the amount of tasks that are downloaded to client - max_concurrent is the amount of tasks that are run at a time (and maybe its infinite-download bug was fixed in the latest release of BOINC): https://github.com/BOINC/boinc/issues/4322 |
Sagittarius Lupus Send message Joined: 12 Apr 11 Posts: 2 Credit: 711,212 RAC: 378 |
Thank you for this thread. I hadn't been able to get any tasks from this project to run for literally years (with a dozen other BOINC projects on this system, some of which also use VBox) until I happened to look here. The advice to limit the camb_boinc2docker application to 8 cores or fewer if no CPU time or output is observed is essential -- it really ought to be pinned somewhere if not made a project default. |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
Unfortunately this seems to be one of those projects (like Rosetta) where nobody who works there is present in the forums, so nothing will get changed. We're on our own. |
Sagittarius Lupus Send message Joined: 12 Apr 11 Posts: 2 Credit: 711,212 RAC: 378 |
How very sad. That's many thousands of work units over the years just left on the table they could have had for the low low price of paying attention. |
Peter Hucker Send message Joined: 5 Jul 11 Posts: 30 Credit: 1,853,885 RAC: 67 |
It's possible they get enough work back and the scientists couldn't deal with any more results anyway. This is the case with GPU work on covid over at WCG (before they broke the server completely for 3 months....) |