Advanced search

Message boards : News : CPU jobs on Linux

Author Message
Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1905
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 48820 - Posted: 5 Feb 2018 | 19:30:28 UTC

Hi, we need more CPUs on Linux to run QM simulations. Anybody can help?

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48821 - Posted: 5 Feb 2018 | 20:05:56 UTC

Do you want to make them more appealing to crunch?? Take the QMML tasks OFF the Boinc CreditNew credit award mechanism and assign them fixed values like you do for the gpu tasks.

biodoc
Send message
Joined: 26 Aug 08
Posts: 104
Credit: 677,870,215
RAC: 1,066,677
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 48823 - Posted: 5 Feb 2018 | 20:47:06 UTC

Sure, I can help.

Matt Kowal
Send message
Joined: 27 May 14
Posts: 1
Credit: 1,188,500
RAC: 0
Level
Ala
Scientific publications
wat
Message 48824 - Posted: 5 Feb 2018 | 21:18:15 UTC
Last modified: 5 Feb 2018 | 21:23:18 UTC

This forum news post was syndicated to your Twitter account, however, the link is broken.

Relevant post: https://twitter.com/gpugrid/status/960604705171808256

The link resolves to https://www.gpugrid.net/extra_arg_utm_source.html

I have reposted your call to the BOINC subreddit

klepel
Send message
Joined: 23 Dec 09
Posts: 149
Credit: 1,903,969,099
RAC: 967,964
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 48825 - Posted: 5 Feb 2018 | 21:45:30 UTC - in response to Message 48821.
Last modified: 5 Feb 2018 | 21:47:43 UTC

Do you want to make them more appealing to crunch?? Take the QMML tasks OFF the Boinc CreditNew credit award mechanism and assign them fixed values like you do for the gpu tasks.

+1

PS Unfortunatelly, they crash my computer with my strongest GPU frequently, so I will not run them on this computer.

John C MacAlister
Send message
Joined: 17 Feb 13
Posts: 179
Credit: 132,357,411
RAC: 3
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 48826 - Posted: 5 Feb 2018 | 22:51:50 UTC - in response to Message 48820.

Unfortunately, I gave up on Linux and run Win 10.


Hi, we need more CPUs on Linux to run QM simulations. Anybody can help?


____________
John

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1905
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 48827 - Posted: 5 Feb 2018 | 23:23:41 UTC - in response to Message 48821.

Do you want to make them more appealing to crunch?? Take the QMML tasks OFF the Boinc CreditNew credit award mechanism and assign them fixed values like you do for the gpu tasks.



We are testing this.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48828 - Posted: 6 Feb 2018 | 0:22:33 UTC - in response to Message 48826.

I've had good luck with the Linux apps up until the recent gpu application errors that started this month. The cpu tasks ran fine.

I and others have voiced our displeasure with the credit awarded for the flops used for the QM tasks in this thread.
New Student and QMML Project

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1905
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 48833 - Posted: 6 Feb 2018 | 9:02:58 UTC - in response to Message 48828.

we are testing different credit systems now.

Toni
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 9 Dec 08
Posts: 623
Credit: 4,273,184
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 48834 - Posted: 6 Feb 2018 | 10:45:39 UTC - in response to Message 48833.
Last modified: 6 Feb 2018 | 10:46:23 UTC

We don't use CreditNew but the previous credit system. In any case, two changes were made yesterday:

* CPU threads are limited to 4 (you should still be able to crunch multiple WUs at once, please check)
* Credits should be in line with other projects'

Let us know.

Profile Bikermatt
Send message
Joined: 8 Apr 10
Posts: 37
Credit: 1,945,860,090
RAC: 1,798,706
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 48836 - Posted: 6 Feb 2018 | 12:54:52 UTC - in response to Message 48834.


* CPU threads are limited to 4 (you should still be able to crunch multiple WUs at once, please check)

Let us know.


Boinc is still assigning all of my 32 threads to one task even though it is only using 4.

Jim1348
Send message
Joined: 28 Jul 12
Posts: 493
Credit: 1,131,859,889
RAC: 18,899
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwat
Message 48837 - Posted: 6 Feb 2018 | 13:29:06 UTC - in response to Message 48820.

Hi, we need more CPUs on Linux to run QM simulations. Anybody can help?

I have three machines on it, but I can run only one work unit at a time on average. That is because when any more start up at once, they error out, as has been discussed before. And I run two cores per work unit for efficiency. But if you could solve the start-up problem, I could run more.

Jim1348
Send message
Joined: 28 Jul 12
Posts: 493
Credit: 1,131,859,889
RAC: 18,899
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwat
Message 48838 - Posted: 6 Feb 2018 | 13:55:26 UTC

I don't see that you have made an announcement on the BOINC forum yet. The Projects section would probably be best.
http://boinc.berkeley.edu/dev/forum_forum.php?id=11

biodoc
Send message
Joined: 26 Aug 08
Posts: 104
Credit: 677,870,215
RAC: 1,066,677
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 48839 - Posted: 6 Feb 2018 | 14:24:00 UTC

I have an Intel 2600K with 8 logical cores. I wanted to reserved 2 cores for feeding 2 GPUs that are running Folding@home. I set the computing preference in boinc to use, at most, 80% of processors (6 cores).

Data on 19 WUs before GPUGrid changes (processor usage varied but 5-6 cores on average I think)

Average run time (sec): 3,129.23
Average CPU time (sec): 16,122.19
Average credit per WU: 228.7
Average WU per day: 27.6
PPD: 6314.8

Data on 15 WUs after GPUGrid changes. Processor usage was 4 cores even though set at 6)

Average run time (sec): 3,566.32
Average CPU time (sec): 14,114.28
Average credit per WU: 819.284
Average WU per day: 24.2
PPD: 19848.5

Summary: Processor usage seems to be maxed out at 4. Run time has increased, CPU time has decreased, WU completed per day has decreased and PPD has increased significantly.

Toni
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 9 Dec 08
Posts: 623
Credit: 4,273,184
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 48840 - Posted: 6 Feb 2018 | 14:57:09 UTC - in response to Message 48836.
Last modified: 6 Feb 2018 | 15:09:56 UTC



Boinc is still assigning all of my 32 threads to one task even though it is only using 4.


Ouch. This should not happen. May be fixed now.

mmonnin
Send message
Joined: 2 Jul 16
Posts: 104
Credit: 120,877,592
RAC: 770,961
Level
Cys
Scientific publications
wat
Message 48842 - Posted: 6 Feb 2018 | 15:50:39 UTC - in response to Message 48837.

Hi, we need more CPUs on Linux to run QM simulations. Anybody can help?

I have three machines on it, but I can run only one work unit at a time on average. That is because when any more start up at once, they error out, as has been discussed before. And I run two cores per work unit for efficiency. But if you could solve the start-up problem, I could run more.


I agree. There were too many issues that had not been resolved. On top of that the credit much worse than even CreditNew. I see the credit has been changed just saying there are reasons for the lack of CPU time.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48845 - Posted: 6 Feb 2018 | 17:48:28 UTC - in response to Message 48834.

Are you sure that you are using an "older" credit mechanism? From the response in the "New Student and QMML" thread from Richard Haselgrove who corrected me in my assumption you might be using an "older" mechanism.

Be careful of your terms. 'CreditNew' has been the default BOINC mechanism since 2010. I suspect this is what GPUGrid is using for these tasks: the support mechanisms for 'even older credit' have been removed from the codebase.

I wonder where you found the older codebase that has been removed that contained the "older" credit award algorithm. If you do in fact have such, I would like access to it. Or have it reinstituted into the BOINC Github codebase.

It would be helpful in persuading the BOINC maintainers that there is in fact a way to return to the older credit algorithm. One of their stated reasons why they said they would not change from CreditNew is that they said they no longer have the original code and can't replicate it.

That said, it looks like the award for QC cpu tasks is much more appealing now.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48847 - Posted: 6 Feb 2018 | 19:21:26 UTC - in response to Message 48839.

Hmmm ... I just ran a new QC task with the supposed new credit. Not seeing any difference.

Run time 3,292.22
CPU time 12,925.47
Validate state Valid
Credit 110.63

I used 4 cores to generate 110 credits for 54 minutes of compute time.

I can use one core to generate 108 credits for 60 minutes of compute time for SETI CPU tasks.

No reason to run these tasks still for me.

STARBASEn
Avatar
Send message
Joined: 17 Feb 09
Posts: 34
Credit: 376,796,362
RAC: 607,160
Level
Asp
Scientific publications
watwatwatwatwat
Message 48850 - Posted: 6 Feb 2018 | 23:50:37 UTC

Since just after mid day yesterday (UTC) I have noticed an increase in credit awarded for the QC WU's. Doing some quick calcs, it appears the increase is about 4.5x of what we were getting. It also appears they are more fixed in value proportional to the size of the WU. My faster machines are getting over 500 credits/hr (4 cores) compute time whereas the slower machines are getting proportionately less/hr and still getting equivalent credit but over a longer period of time than the faster ones.

Well, at least my avg credit per day will not be taking as much of a hit per day as it has been without Linux GPU WU's :).

So far, I have 5 machines with 4 cores each running the QC project and will add one more when I install the memory upgrade on one of the headless systems.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48851 - Posted: 7 Feb 2018 | 0:28:50 UTC - in response to Message 48850.

I'm not sure the higher credit is not due to the larger molecule size in the latest tasks that Dominik explained to me here.

Trotador
Send message
Joined: 25 Mar 12
Posts: 86
Credit: 1,104,668,560
RAC: 332,675
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 48857 - Posted: 7 Feb 2018 | 6:26:25 UTC

No higher credits for the tasks I've crunched yesterday and today.

I stop and will try to understand what is happening.

NUCCpod_NAPTIMELABS_01
Send message
Joined: 18 Aug 17
Posts: 3
Credit: 1,730,296
RAC: 136,808
Level
Ala
Scientific publications
wat
Message 48858 - Posted: 7 Feb 2018 | 6:33:41 UTC - in response to Message 48837.

I've also been having trouble with work units erroring out in this way. I have over 150 cpu cores spread out across a variety of machines, all under linux. Just a few moments ago I attempted to attach to GPUGRID only to have computational error after computational error. I hope this is fixed shortly as I would love to have GPUGRID as one of my default projects.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48859 - Posted: 7 Feb 2018 | 7:46:25 UTC

Yes, credit is doing something very strange. I got 111 credits for 3292 seconds of cpu time.

klepel got 1362 credits for the same time.

Run time 3,262.28
CPU time 12,834.34
Validate state Valid
Credit 1,361.79

Task 13118994

kain
Send message
Joined: 3 Sep 14
Posts: 112
Credit: 143,430,213
RAC: 357,589
Level
Cys
Scientific publications
watwatwatwat
Message 48862 - Posted: 7 Feb 2018 | 12:23:54 UTC - in response to Message 48847.


I used 4 cores to generate 110 credits for 54 minutes of compute time.

I can use one core to generate 108 credits for 60 minutes of compute time for SETI CPU tasks.

No reason to run these tasks still for me.


Are you seriously comparing SETI to GPUGRID?!

mmonnin
Send message
Joined: 2 Jul 16
Posts: 104
Credit: 120,877,592
RAC: 770,961
Level
Cys
Scientific publications
wat
Message 48863 - Posted: 7 Feb 2018 | 12:53:02 UTC - in response to Message 48862.


I used 4 cores to generate 110 credits for 54 minutes of compute time.

I can use one core to generate 108 credits for 60 minutes of compute time for SETI CPU tasks.

No reason to run these tasks still for me.


Are you seriously comparing SETI to GPUGRID?!


CreditNew is CreditNew.

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 297
Credit: 1,657,319,656
RAC: 5,267,556
Level
His
Scientific publications
watwat
Message 48864 - Posted: 7 Feb 2018 | 13:33:40 UTC - in response to Message 48863.


I used 4 cores to generate 110 credits for 54 minutes of compute time.

I can use one core to generate 108 credits for 60 minutes of compute time for SETI CPU tasks.

No reason to run these tasks still for me.


Are you seriously comparing SETI to GPUGRID?!


CreditNew is CreditNew.

You cannot compare one project's credit to another.

computezrmle
Send message
Joined: 10 Jun 13
Posts: 5
Credit: 10,648,441
RAC: 175,627
Level
Pro
Scientific publications
wat
Message 48866 - Posted: 7 Feb 2018 | 16:54:10 UTC - in response to Message 48864.

You cannot compare one project's credit to another.

At least it should be comparable as described in the BOINC documentation.
See: http://boinc.berkeley.edu/trac/wiki/CreditNew#Cross-projectversionnormalization

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48867 - Posted: 7 Feb 2018 | 17:01:28 UTC - in response to Message 48864.


I used 4 cores to generate 110 credits for 54 minutes of compute time.

I can use one core to generate 108 credits for 60 minutes of compute time for SETI CPU tasks.

No reason to run these tasks still for me.


Are you seriously comparing SETI to GPUGRID?!


CreditNew is CreditNew.

You cannot compare one project's credit to another.

One of the stated objectives of CreditNew it to make credit the same across all projects for the same amount of cobblestones used to compute.

mmonnin
Send message
Joined: 2 Jul 16
Posts: 104
Credit: 120,877,592
RAC: 770,961
Level
Cys
Scientific publications
wat
Message 48872 - Posted: 7 Feb 2018 | 18:35:25 UTC

Thus my comment about CreditNew. CPU projects that have higher or lower then typical RAC are most likely using something else besides CreditNew like a fixed credit or another algorithm.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48873 - Posted: 7 Feb 2018 | 18:46:31 UTC - in response to Message 48872.

Yes I understood your post and sentiment. My post was directed at the other poster's incredulous comment.

This project itself utilize both mechanisms. CreditNew for cpu tasks and fixed credit awards for gpu tasks.

As far as I have been able to find, that is unique among projects. Usually it is either/or not both.

mmonnin
Send message
Joined: 2 Jul 16
Posts: 104
Credit: 120,877,592
RAC: 770,961
Level
Cys
Scientific publications
wat
Message 48876 - Posted: 7 Feb 2018 | 22:11:38 UTC - in response to Message 48873.

Yes I understood your post and sentiment. My post was directed at the other poster's incredulous comment.

This project itself utilize both mechanisms. CreditNew for cpu tasks and fixed credit awards for gpu tasks.

As far as I have been able to find, that is unique among projects. Usually it is either/or not both.


I wasn't referencing you as I didn't quote you. ;)

I agree.

Profile Michael H.W. Weber
Send message
Joined: 9 Feb 16
Posts: 32
Credit: 281,590,971
RAC: 284,651
Level
Asn
Scientific publications
watwat
Message 48893 - Posted: 10 Feb 2018 | 15:13:42 UTC
Last modified: 10 Feb 2018 | 15:18:45 UTC

Would you please list the QC project progress on the server status page as well:

http://www.gpugrid.net/server_status.php

Thanks.

Another issue is that your app does not dynamically allocate CPU cores according to the BOINC settings. Instead it claims all physically present cores. That is a major problem when trying to run computations on the GPU as well because to do so, the GPU project automatically (or I manually) reserve(s) one CPU core per GPU task.
Example: When the BOINC manager is set to use 7 of the 8 cores to do CPU computations, your CPU client grabs all 8 cores (or since recently 2x 4 cores).
That is not acceptable.
Please fix this to attract more people to donate CPU cycles to your project.

Michael.
____________
President of Rechenkraft.net - Germany's first and largest distributed computing organization.

Profile SMTB1963
Avatar
Send message
Joined: 27 Jun 10
Posts: 37
Credit: 214,222,621
RAC: 626,703
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwat
Message 48894 - Posted: 10 Feb 2018 | 17:48:45 UTC - in response to Message 48893.

Another issue is that your app does not dynamically allocate CPU cores according to the BOINC settings. Instead it claims all physically present cores.


I'm seeing this behavior as well. On my Ryzen 1700X system with 2 GPUs, these WUs basically take over all CPUs and throw the GPUs into "Waiting to run".

I suppose one could set max_concurrent in an app_config.xml to fix this...what would be the proper app name to use?

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 80
Credit: 7,566,813
RAC: 152,251
Level
Ser
Scientific publications
wat
Message 48895 - Posted: 10 Feb 2018 | 17:54:26 UTC - in response to Message 48894.

QC is the proper app name. This is how I limit QC to 2 threads per task.

<app>
<name>QC</name>
<max_concurrent>1</max_concurrent>
</app>
<app_version>
<app_name>QC</app_name>
<plan_class>mt</plan_class>
<avg_ncpus>2.000000</avg_ncpus>
<cmdline>--nthreads 2</cmdline>


I also just run 1 task at a time to avoid the starting two tasks at the same time flaw in the application.

Post to thread

Message boards : News : CPU jobs on Linux