Server code update and fix to GPU work scheduling

log in

Advanced search

Message boards : News : Server code update and fix to GPU work scheduling

Author Message
Teemu Mannermaa
Project administrator
Project developer
Project tester
Send message
Joined: 20 Apr 11
Posts: 308
Credit: 366,951,821
RAC: 381,805
Message 3709 - Posted: 10 Sep 2012, 4:17:53 UTC

Just finished an update to the latest BOINC Server code that brought a lot of fixes and improvements from BOINC devs in last 6 months. These included web feature changes as well as various backend process changes. (And before you ask, no, credit system is still the same and no granting adjustments have been done there.)

Additionally, since the code refresh didn't seem to fix it, I debugged the "only 1 GPU work unit sent by scheduler" problem and hopefully got it fixed. It seems the matchmaker scoring scheduling algorithm we switched to some time ago doesn't work correctly for GPU work request. (Specifically the new resource based work requests of new BOINC client versions.) I switched us back to the old trusty array scheduling algorithm and now the GPU work requests are been fulfilled fully by our scheduler.

As always, please report any oddities you might encounter and thanks for crunching!

mikey
Avatar
Send message
Joined: 22 Jun 11
Posts: 1375
Credit: 1,000,866,048
RAC: 0
Message 3713 - Posted: 10 Sep 2012, 11:17:44 UTC - in response to Message 3709.

Just finished an update to the latest BOINC Server code that brought a lot of fixes and improvements from BOINC devs in last 6 months. These included web feature changes as well as various backend process changes. (And before you ask, no, credit system is still the same and no granting adjustments have been done there.)

Additionally, since the code refresh didn't seem to fix it, I debugged the "only 1 GPU work unit sent by scheduler" problem and hopefully got it fixed. It seems the matchmaker scoring scheduling algorithm we switched to some time ago doesn't work correctly for GPU work request. (Specifically the new resource based work requests of new BOINC client versions.) I switched us back to the old trusty array scheduling algorithm and now the GPU work requests are been fulfilled fully by our scheduler.

As always, please report any oddities you might encounter and thanks for crunching!


Thank you very much, I hope it fixes my problem too!

tuankiet65
Send message
Joined: 30 May 12
Posts: 1
Credit: 33,112
RAC: 0
Message 3717 - Posted: 10 Sep 2012, 12:40:35 UTC - in response to Message 3709.

thanks. That's why when BOINC on my computer get GPU work, there is only 1 work sent

Profile dude8724
Avatar
Send message
Joined: 30 Dec 11
Posts: 10
Credit: 768,160
RAC: 0
Message 3729 - Posted: 12 Sep 2012, 4:14:21 UTC

I received 10 cuda tasks, but for some reason they never seem to run. My last single cuda task expired after sitting in my queue for over a week. Moo! Wrapper is the only project that even seems to download cuda task for my dual GeForce GTX 550 Ti computer since I upgraded to BOINC 7.0.28.

The top of my BOINC Event Log:


9/8/2012 5:46:01 PM | | No config file found - using defaults
9/8/2012 5:46:01 PM | | Starting BOINC client version 7.0.28 for windows_x86_64
9/8/2012 5:46:01 PM | | log flags: file_xfer, sched_ops, task
9/8/2012 5:46:01 PM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
9/8/2012 5:46:01 PM | | Data directory: C:\ProgramData\BOINC
9/8/2012 5:46:01 PM | | Running under account Dude
9/8/2012 5:46:01 PM | | Processor: 4 AuthenticAMD AMD FX(tm)-4100 Quad-Core Processor [Family 21 Model 1 Stepping 2]
9/8/2012 5:46:01 PM | | Processor: 2.00 MB cache
9/8/2012 5:46:01 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm svm sse4a osvw ibs xop skinit wdt lwp fma4 page1gb rdtscp
9/8/2012 5:46:01 PM | | OS: Microsoft Windows 7: Home Premium x64 Edition, Service Pack 1, (06.01.7601.00)
9/8/2012 5:46:01 PM | | Memory: 16.00 GB physical, 31.99 GB virtual
9/8/2012 5:46:01 PM | | Disk: 494.73 GB total, 375.49 GB free
9/8/2012 5:46:01 PM | | Local time is UTC -4 hours
9/8/2012 5:46:01 PM | | VirtualBox version: 4.1.8
9/8/2012 5:46:01 PM | | NVIDIA GPU 0: GeForce GTX 550 Ti (driver version 301.42, CUDA version 4.20, compute capability 2.1, 1024MB, 817MB available, 703 GFLOPS peak)
9/8/2012 5:46:01 PM | | NVIDIA GPU 1: GeForce GTX 550 Ti (driver version 301.42, CUDA version 4.20, compute capability 2.1, 1024MB, 817MB available, 703 GFLOPS peak)
9/8/2012 5:46:01 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 550 Ti (driver version 301.42, device version OpenCL 1.1 CUDA, 1024MB, 817MB available)
9/8/2012 5:46:01 PM | | OpenCL: NVIDIA GPU 1: GeForce GTX 550 Ti (driver version 301.42, device version OpenCL 1.1 CUDA, 1024MB, 817MB available)
9/8/2012 5:46:01 PM | Constellation | URL http://aerospaceresearch.net/constellation/; Computer ID 7006; resource share 500
9/8/2012 5:46:01 PM | rosetta@home | URL http://boinc.bakerlab.org/rosetta/; Computer ID 1503442; resource share 150
9/8/2012 5:46:01 PM | DistrRTgen | URL http://boinc.freerainbowtables.com/distrrtgen/; Computer ID 31325; resource share 500
9/8/2012 5:46:01 PM | Poem@Home | URL http://boinc.fzk.de/poem/; Computer ID 110049; resource share 100
9/8/2012 5:46:01 PM | Leiden Classical | URL http://boinc.gorlaeus.net/; Computer ID 92770; resource share 100
9/8/2012 5:46:01 PM | Collatz Conjecture | URL http://boinc.thesonntags.com/collatz/; Computer ID 90139; resource share 200
9/8/2012 5:46:01 PM | The Lattice Project | URL http://boinc.umiacs.umd.edu/; Computer ID 73419; resource share 100
9/8/2012 5:46:01 PM | boincsimap | URL http://boincsimap.org/boincsimap/; Computer ID 219448; resource share 100
9/8/2012 5:46:01 PM | BURP | URL http://burp.renderfarming.net/; Computer ID 52919; resource share 500
9/8/2012 5:46:01 PM | climateprediction.net | URL http://climateprediction.net/; Computer ID 1185692; resource share 200
9/8/2012 5:46:01 PM | Docking | URL http://docking.cis.udel.edu/; Computer ID 107264; resource share 100
9/8/2012 5:46:01 PM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 4309125; resource share 200
9/8/2012 5:46:01 PM | Test4Theory@Home | URL http://lhcathome2.cern.ch/test4theory/; Computer ID 14461; resource share 500
9/8/2012 5:46:01 PM | LHC@home 1.0 | URL http://lhcathomeclassic.cern.ch/sixtrack/; Computer ID 9948039; resource share 100
9/8/2012 5:46:01 PM | Milkyway@Home | URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 355680; resource share 200
9/8/2012 5:46:01 PM | MindModeling@Beta | URL http://MindModeling.org/beta/; Computer ID 19162; resource share 100
9/8/2012 5:46:01 PM | Moo! Wrapper | URL http://moowrap.net/; Computer ID 7910; resource share 200
9/8/2012 5:46:01 PM | orbit@home | URL http://orbit.psi.edu/oah/; Computer ID 82049; resource share 100
9/8/2012 5:46:01 PM | Pirates@Home | URL http://pirates.spy-hill.net/; Computer ID 71799; resource share 100
9/8/2012 5:46:01 PM | QMC@HOME | URL http://qah.uni-muenster.de/; Computer ID 203296; resource share 100
9/8/2012 5:46:01 PM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 6293163; resource share 1000
9/8/2012 5:46:01 PM | SETI@home Beta Test | URL http://setiweb.ssl.berkeley.edu/beta/; Computer ID 55373; resource share 100
9/8/2012 5:46:01 PM | Spinhenge@home | URL http://spin.fh-bielefeld.de/; Computer ID 220826; resource share 100
9/8/2012 5:46:01 PM | Cosmology@Home | URL http://www.cosmologyathome.org/; Computer ID 151072; resource share 100
9/8/2012 5:46:01 PM | Enigma@Home | URL http://www.enigmaathome.net/; Computer ID 71412; resource share 100
9/8/2012 5:46:01 PM | FreeHAL@home | URL http://www.freehal.net/freehal_at_home/; Computer ID 72013; resource share 100
9/8/2012 5:46:01 PM | GPUGRID | URL http://www.gpugrid.net/; Computer ID 115985; resource share 200
9/8/2012 5:46:01 PM | PrimeGrid | URL http://www.primegrid.com/; Computer ID 240499; resource share 100
9/8/2012 5:46:01 PM | yoyo@home | URL http://www.rechenkraft.net/yoyo/; Computer ID 63254; resource share 100
9/8/2012 5:46:01 PM | uFluids | URL http://www.ufluids.net/; Computer ID 135801; resource share 100
9/8/2012 5:46:01 PM | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 1830670; resource share 200
9/8/2012 5:46:01 PM | Moo! Wrapper | General prefs: from Moo! Wrapper (last modified 09-Aug-2012 19:33:33)
9/8/2012 5:46:01 PM | Moo! Wrapper | Host location: none
9/8/2012 5:46:01 PM | Moo! Wrapper | General prefs: using your defaults
9/8/2012 5:46:01 PM | | Reading preferences override file
9/8/2012 5:46:01 PM | | Preferences:
9/8/2012 5:46:01 PM | | max memory usage when active: 12285.93MB
9/8/2012 5:46:01 PM | | max memory usage when idle: 16381.24MB
9/8/2012 5:46:01 PM | | max disk usage: 20.00GB
9/8/2012 5:46:01 PM | | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
9/8/2012 5:46:01 PM | | Not using a proxy
9/8/2012 5:46:02 PM | | Contacting account manager at http://bam.boincstats.com/

Most of the other GPU task aborted right after the upgrade, but now I don't seem to get any. I did recently change the allowed queue from minimum 0.1 days and 0.0 extra to 0.9 days minimum and 0.1 extra after saying something about the lack of GPU work for my computer since BOINC 7.0.x upgrade.

I did see something about an app settings file for fixing dual ATI cards in another post in the Moo! Wrapper message board.
____________
David - KC2WUF

mikey
Avatar
Send message
Joined: 22 Jun 11
Posts: 1375
Credit: 1,000,866,048
RAC: 0
Message 3734 - Posted: 12 Sep 2012, 11:08:58 UTC - in response to Message 3729.

I received 10 cuda tasks, but for some reason they never seem to run. My last single cuda task expired after sitting in my queue for over a week. Moo! Wrapper is the only project that even seems to download cuda task for my dual GeForce GTX 550 Ti computer since I upgraded to BOINC 7.0.28.

I did see something about an app settings file for fixing dual ATI cards in another post in the Moo! Wrapper message board.


And that last part is the ONLY way one of my pc's can run Moo on it! I have NO IDEA why, it is similar to all the rest of my pc's but it just won't run Moo without that app_info.xml file! Prior to that I got the exact same thing you did, it would download units but never run a single one of them! I even detached from all other gpu projects on that machine and Moo STILL wouldn't run!! I have AMD cards in that machine so it is NOT just AMD or just Nvidia cards!! Right now the machine is crunching thru the units using the dual AMD cards but it does give me a weird error message every day:
9/12/2012 7:06:20 AM | Moo! Wrapper | Not requesting tasks: project is not highest priority

I have no idea how that affects anything as I DO still get units, just not right now. In the end it IS working now for me so I am not messing with it too much!

Cristi
Send message
Joined: 15 Nov 11
Posts: 2
Credit: 28,893,852
RAC: 2
Message 3738 - Posted: 12 Sep 2012, 23:39:22 UTC

Does this fix the AMD radeon HD 7xxx series problem (computation error after around 20 seconds)?

Teemu Mannermaa
Project administrator
Project developer
Project tester
Send message
Joined: 20 Apr 11
Posts: 308
Credit: 366,951,821
RAC: 381,805
Message 3740 - Posted: 13 Sep 2012, 4:03:43 UTC - in response to Message 3738.

Does this fix the AMD radeon HD 7xxx series problem (computation error after around 20 seconds)?


Unfortunately no, that needs a new OpenCL based application since AMD decided to suddenly drop support for CAL. D.net is working on that but no ETA yet when that client passes internal testing.

-w

Teemu Mannermaa
Project administrator
Project developer
Project tester
Send message
Joined: 20 Apr 11
Posts: 308
Credit: 366,951,821
RAC: 381,805
Message 3741 - Posted: 13 Sep 2012, 4:10:57 UTC - in response to Message 3729.

I received 10 cuda tasks, but for some reason they never seem to run. My last single cuda task expired after sitting in my queue for over a week. Moo! Wrapper is the only project that even seems to download cuda task for my dual GeForce GTX 550 Ti computer since I upgraded to BOINC 7.0.28.


Yeah, it seems BOINC Client v7 has some kind of bug when dealing with tasks that require more than 1 GPU. :( To speed things up, you should try to report this to the BOINC devs in the hope they can fix it. Second option would be to carefully (crunch/report every task since the queues will get reset) downgrade back to the latest v6 release that works with multi-GPU setups. Third one is using the app settings file since it seems anonymous platforms are not suffering from the bug.

-w

Profile MatthiasLeimbach
Send message
Joined: 13 Mar 12
Posts: 1
Credit: 8,729,534
RAC: 0
Message 3938 - Posted: 7 Oct 2012, 8:01:29 UTC

Last week I have had severe raised heat running Distributed.net Client 1.03, had to bring Boinc down to 15 % cpu time usage, the World Community Grid units remain normal

Matthias

Message boards : News : Server code update and fix to GPU work scheduling


Main page · Your account · Message boards


Copyright © 2011-2014 Moo! Wrapper Project