High performance Linux clients at SETI

Message boards : Number crunching : High performance Linux clients at SETI
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · 8 · 9 . . . 21 · Next

AuthorMessage
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3379
Credit: 173,935,112
RAC: 790,593
United States
Message 1986281 - Posted: 21 Mar 2019, 2:43:38 UTC - in response to Message 1986112.  

I just got it going, nothing yet. I tried reducing the cache and hitting update, and it didn't jump. I'll give it 6 hours, if it hasn't jumped by then it probably isn't going to jump.
So, what's the record for fixing the longest standing BOINC BUG? This has to be close.
Well, it isn't fixed yet, but I have created #3064 to prompt review by other coders and hopefully implementation.

No, this isn't the longest-lived complaint, by a long way. I'd nominate #41 (July 2007) for that honor, but it would take longer than half an hour to fix.


I wonder if a workaround for #41 would be the setting we have where if the system gets busier than XX% Boinc suspends? Or else basically what he wants is instead of having all of boinc suspend, only part of boinc would suspend.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1986281 · Report as offensive     Reply Quote
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4745
Credit: 509,707,286
RAC: 1,238,136
United States
Message 1986607 - Posted: 22 Mar 2019, 17:39:57 UTC

I just installed a Fresh copy of Ubuntu 18.04.2 to make sure everything was as it appeared. A fresh install of 18.04.2 will give you Only libcurl3-gnutis, it won't install either libcurl3 or libcurl4....Only libcurl3-gnutis. Running BOINC 7.14.2 compiled with Ubuntu 14.04.1 will require both libwebkitgtk-1.0-0 and libcurl3 when run in 18.04.2, just as with BOINC 7.4.44 and 7.8.3.
Running BOINC 7.14.2 compiled with Ubuntu 15.04 doesn't require any dependencies when run in Ubuntu 15.04 and above, including 18.04.2 and 18.10.
Just as with about everything else, to run BOINC you must have at least as high as what was used to compile the App or above. This Means the BOINC compiled in 15.04 Will Not Work in 14.04.x, it also means Anything compiled in Ubuntu 18.04.x will Not Work with any lower version of Ubuntu, this includes a certain BOINC 7.15.0 floating around.
So, the next All-In-One will have the BOINC compiled in 15.04 as the Default, and the version for Ubuntu 14.04 in a separate folder for Manual installation. The 14.04 version will work in 14.04 and above with 2 dependencies, the Default version will work in 15.04 and above without dependencies. Anything compiled in 18.04 will require 18.04 to run, that's usually how it works.
ID: 1986607 · Report as offensive     Reply Quote
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4745
Credit: 509,707,286
RAC: 1,238,136
United States
Message 1986648 - Posted: 22 Mar 2019, 22:34:01 UTC

The New All-In-One is in place, at the same place, http://www.arkayn.us/lunatics/BOINC.7z
If you are starting from Scratch, unpack the BOINC folder to your Home folder, open the folder and double-click on boincmgr. Then add SETI@home to BOINC. Files will be downloaded and run with the CUDA90 App. If something doesn't work make sure the Execute bits are set, and then check the Dependencies by opening a Terminal window, typing ldd and a space, then dragging and dropping the App into the window. Bring the Terminal window to the front, hit enter, and look for missing dependencies.

If you already have a BOINC folder in Home, Stop BOINC, unpack the new BOINC folder to Downloads, copy the desired files from the downloaded BOINC and paste them into the BOINC in Home. The Five BOINC files are;
boinc
boinccmd
boincmgr
boincscr
switcher
BOINC files for the older systems are in the docs folder.
If you want to use the run_client & run_manager scripts, you need to add your user folder name to the location replacing 'user'.
The download is a little larger than last time, but it seems to be downloading very slowly at present. Don't know, it uploaded pretty quick, as usual.
ID: 1986648 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 9266
Credit: 826,527,033
RAC: 1,850,946
United States
Message 1986649 - Posted: 22 Mar 2019, 22:55:48 UTC

Thanks TBar. We appreciate the work you put into developing the All-in-One package.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1986649 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3379
Credit: 173,935,112
RAC: 790,593
United States
Message 1986663 - Posted: 23 Mar 2019, 0:13:12 UTC - in response to Message 1986649.  

Thanks TBar. We appreciate the work you put into developing the All-in-One package.


And supporting it leading you into exhaustion where you wisely stopped burning yourself out.

We thank you!

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1986663 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5146
Credit: 400,357,991
RAC: 1,536,230
United States
Message 1986674 - Posted: 23 Mar 2019, 1:06:07 UTC - in response to Message 1986649.  

Thanks TBar. We appreciate the work you put into developing the All-in-One package.


+2
ID: 1986674 · Report as offensive     Reply Quote
Lane42

Send message
Joined: 17 May 99
Posts: 55
Credit: 205,282,410
RAC: 466,124
United States
Message 1986681 - Posted: 23 Mar 2019, 2:16:36 UTC

Great job TBar, Thanks
ID: 1986681 · Report as offensive     Reply Quote
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 7890
Credit: 475,202,195
RAC: 648,892
Panama
Message 1986682 - Posted: 23 Mar 2019, 2:30:51 UTC - in response to Message 1986674.  

Thanks TBar. We appreciate the work you put into developing the All-in-One package.


+2

+3
ID: 1986682 · Report as offensive     Reply Quote
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 146
Credit: 24,198,539
RAC: 95,975
Australia
Message 1986708 - Posted: 23 Mar 2019, 8:14:26 UTC - in response to Message 1986682.  

+4
ID: 1986708 · Report as offensive     Reply Quote
Profile Dr.Diesel Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 14 May 99
Posts: 41
Credit: 104,516,253
RAC: 86,354
United States
Message 1986874 - Posted: 24 Mar 2019, 10:23:28 UTC - in response to Message 1986708.  

Many thanks TBar.
ID: 1986874 · Report as offensive     Reply Quote
Joe Januzzi
Volunteer tester
Avatar

Send message
Joined: 13 Apr 03
Posts: 45
Credit: 263,302,026
RAC: 129,638
United States
Message 1986916 - Posted: 24 Mar 2019, 16:51:13 UTC - in response to Message 1986874.  

+5

Real Join Date:
Joe Januzzi (ID 253343) 29 Sep 1999, 22:30:36 UTC
Try to learn something new everyday.
ID: 1986916 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 13022
Credit: 142,474,845
RAC: 179,270
United Kingdom
Message 1990232 - Posted: 16 Apr 2019, 9:50:56 UTC - in response to Message 1985956.  

So, if you put a condition round the scroll command, like

                    if (m_pListPane->GetItemCount() <= m_pListPane->GetCountPerPage()) {
                        m_pListPane->EnsureVisible(iDocCount - 1);
                    }
it might not jump when you don't want it to jump. It works in Windows - could somebody test it in Linux, please?
Pleased to report that this patch has been accepted by the scrutineers and will be incorporated in the next client/manager release - probably 7.16, whenever that happens: there is still one showstopper bug which hasn't been fixed, and other items are still under discussion.
ID: 1990232 · Report as offensive     Reply Quote
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4745
Credit: 509,707,286
RAC: 1,238,136
United States
Message 1990239 - Posted: 16 Apr 2019, 10:41:58 UTC - in response to Message 1986648.  

The New All-In-One is in place, at the same place, http://www.arkayn.us/lunatics/BOINC.7z
If you are starting from Scratch, unpack the BOINC folder to your Home folder, open the folder and double-click on boincmgr. Then add SETI@home to BOINC. Files will be downloaded and run with the CUDA90 App. If something doesn't work make sure the Execute bits are set, and then check the Dependencies by opening a Terminal window, typing ldd and a space, then dragging and dropping the App into the window. Bring the Terminal window to the front, hit enter, and look for missing dependencies.

If you already have a BOINC folder in Home, Stop BOINC, unpack the new BOINC folder to Downloads, copy the desired files from the downloaded BOINC and paste them into the BOINC in Home. The Five BOINC files are;
boinc
boinccmd
boincmgr
boincscr
switcher
BOINC files for the older systems are in the docs folder.
If you want to use the run_client & run_manager scripts, you need to add your user folder name to the location replacing 'user'.
The download is a little larger than last time, but it seems to be downloading very slowly at present. Don't know, it uploaded pretty quick, as usual.
A New version has been uploaded to the same location.
New in this version is an upgrade to 0.98b1 featuring numerous improvements to compatibility and speed. Most notable is much better handling of the Arecibo files producing fewer inconclusive results. All Users should update to this version to reduce the number of repeated Validation attempts.

The default versions are BOINC 7.14.2 requiring at least Ubuntu 15.04 or higher, versions for 14.04 & 12.04 are included if needed. These BOINC Apps have the fix which Stops the jumping Tasks & Transfers Pages. The default CUDA App is CUDA 9.0 which will work with the Repository drivers from 14.04 through 18.10. The CUDA 10.1 App is included for those able to update to a CUDA 10.1 driver. ReadMe files are included. This version also features much lower VRAM requirements meaning your 2 GB GPUs can be used for limited other uses while crunching, use NVIDIA XServer Settings or nvidia-smi -l to view VRAM usage.
ID: 1990239 · Report as offensive     Reply Quote
Profile -= Vyper =-
Volunteer tester
Avatar

Send message
Joined: 5 Sep 99
Posts: 1555
Credit: 799,796,412
RAC: 1,060,340
Sweden
Message 1990244 - Posted: 16 Apr 2019, 12:38:32 UTC - in response to Message 1990239.  

The New All-In-One is in place, at the same place


Thank you!!
It seems like it works on the 750Ti aswell.
Shaving of a minute of the time, 24% faster..

Well done lads!!

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group
ID: 1990244 · Report as offensive     Reply Quote
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 223
Credit: 2,700,812,372
RAC: 3,416,585
France
Message 1990251 - Posted: 16 Apr 2019, 16:36:16 UTC

Thanks Petri, TBar and everyone involved in this masterpiece ! :)
ID: 1990251 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 9266
Credit: 826,527,033
RAC: 1,850,946
United States
Message 1990252 - Posted: 16 Apr 2019, 16:38:08 UTC

Thanks Petri for such a nice application. Cheers.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1990252 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 9266
Credit: 826,527,033
RAC: 1,850,946
United States
Message 1990257 - Posted: 16 Apr 2019, 17:03:27 UTC
Last modified: 16 Apr 2019, 17:15:29 UTC

Still waiting for tasks finished with the new app to report so I can inspect the stderr.txt. From the readme it looks like the default unroll is again equal to the number of SM's on the card. But I saw interesting things with unroll =1 on my beta test. So to answer a question, I think I will run the new app through the benchmark again with the default unroll parameter against the new app with unroll=1 to see what if any improvements are gained.

[Edit] I see now that I was wrong. The default unroll is one now with new app. So no need to test in the benchmark. It still would be interesting to compile a version with the old unroll method to get an app to benchmark against this new version since TBar had included the lovely source code for the applications.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1990257 · Report as offensive     Reply Quote
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 1604
Credit: 617,553,324
RAC: 2,951,364
United States
Message 1990271 - Posted: 16 Apr 2019, 17:57:07 UTC - in response to Message 1990257.  

yeah I saw the same. I loaded up the 418 drivers and new app on my 6-GPU system. but initial impressions were that it was ever so slightly slower than the beta version of v0.98 that petri had me testing (which used a normal unroll value equal to the SM/CU count). if there's any slowdown (vs 0.98 beta), it's probably due to features added to make it more compatible with older cards. that's not a bad thing, just might be a little slower on the newer cards if that's the case.

I hope petri can chime in about something else though. Why the change from CUDA 10 to 10.1? was there any specific reason for this? From what i remember, performance was better with 410 drivers (vs 415), so i always stayed there. but having cuda 10.1 requires the 418+ drivers now, which may have the same slowdown in performance that 415 had.

I'll do the same and run my suite of benchmarks to directly compare v0.98 beta (cuda10) to v0.98b1 (cuda10.1) with 418 drivers.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1990271 · Report as offensive     Reply Quote
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 223
Credit: 2,700,812,372
RAC: 3,416,585
France
Message 1990280 - Posted: 16 Apr 2019, 18:44:53 UTC

Unroll always set to 1 ?
No way to change this value in the <cmdline> line ?

GTX 1050 Ti have 4 Gb of RAM but sometimes you can fill them with Xorg/compiz (with two 2K monitor) and some apps like Mathematica.
I used to change the unroll value on the fly according to the remaning free gpu ram.

Does it mean the binary require a fixed amount of ram, whatever the number of sm is used ?
If yes, it's a good news as I don't need anymore to monitor the free ram remaining !
Tell me please !
Thanks
ID: 1990280 · Report as offensive     Reply Quote
JohnDK Crowdfunding Project Donor*Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 28 May 00
Posts: 1109
Credit: 295,862,706
RAC: 436,338
Denmark
Message 1990281 - Posted: 16 Apr 2019, 18:50:22 UTC

From the readme

4) With this version -unroll autotune and Blocking Sync is set by default. Override Autotune with the cmdline -unroll N, Blocking Sync override is -nobs
ID: 1990281 · Report as offensive     Reply Quote
Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · 8 · 9 . . . 21 · Next

Message boards : Number crunching : High performance Linux clients at SETI


 
©2019 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.