Extremely slow computation for some tasks

Questions and Answers : Unix/Linux : Extremely slow computation for some tasks
Message board moderation

To post messages, you must log in.

AuthorMessage
brain

Send message
Joined: 4 Sep 16
Posts: 5
Credit: 382
RAC: 0
Sweden
Message 2011094 - Posted: 7 Sep 2019, 16:57:25 UTC

My computer

I've forced the cores to run at 1.3GHz because of cooling problems. I've completed three tasks in about 3-4 hours each and one completed in about a minute. These four tasks have been named like "05se19aa.23089.9883.7.34.148" etc (which may be a hint to the solution)

Now the client has downloaded new tasks which are called "blcXX_2bit_guppi_xxxx" and the computation is INSANELY slow. Take a look at this screenshot.

If it continues at this rate, the slowest tasks won't be finished within A WEEK. Something must be terribly wrong, right?
ID: 2011094 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 10998
Credit: 1,108,885,200
RAC: 1,666,661
United States
Message 2011104 - Posted: 7 Sep 2019, 18:43:00 UTC - in response to Message 2011094.  
Last modified: 7 Sep 2019, 18:44:07 UTC

The 05 task is a new Arecibo file. The type is a standard AR of around .42-.44 and would normally run on a 3Ghz client in around 2-4 hours on the cpu. The BLC task is a Breakthrough Listen task from the Green Bank Telescope. All BLC tasks are Very Low Angle Range and take longer to crunch than Arecibo standard tasks but not as long as Arecibo VLAR tasks. Your slower cpu will probably take longer, maybe as much as 8 hours.

The other thing to note is that your host is brand new and has not been crunching long enough for the client and scheduler to make good estimates of the time for completion. So you can ignore the estimated times to complete for now. Once the host has crunched and reported 11 validated tasks, and not either an overflow task or one with too much radar blanking, the host will develop an APR (average processing rate) that you can see in the Details page of the host and the estimated time for completion will become accurate. It needs to do this for every application on the host.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011104 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15060
Credit: 4,313,585
RAC: 2,239
Netherlands
Message 2011106 - Posted: 7 Sep 2019, 18:45:55 UTC - in response to Message 2011094.  

You're using an anonymous platform application for calculations. Can you tell which one that is and where you got it from?
Have you tried the default Linux application(s)?
ID: 2011106 · Report as offensive
brain

Send message
Joined: 4 Sep 16
Posts: 5
Credit: 382
RAC: 0
Sweden
Message 2011119 - Posted: 7 Sep 2019, 19:54:05 UTC

First, thank you both for the replies, it's appreciated!

Keith, my concern is not the "remaining (estimated)" number, I know it can be "a bit" off. =) My concern is the actual progress of the computation. For example, as I'm writing this, I have a task which has been running with 100% CPU (1.3 GHz) for 3h 37 minutes and with a progress of 0.444%. At that pace it will be finished with that single task in about 34 days. Something is not right....

Ageless; I've installed the all software from the official Debian repos. See below for version information

boinc-app-seti 8.00~svn4035-1
boinc-client 7.16.1+dfsg-2
boinc-client-nvidia-cuda 7.16.1+dfsg-2
boinc-client-opencl 7.16.1+dfsg-2
boinc-manager 7.16.1+dfsg-2
libboinc-app7:amd64 7.16.1+dfsg-2
libboinc7:amd64 7.16.1+dfsg-2
ID: 2011119 · Report as offensive
brain

Send message
Joined: 4 Sep 16
Posts: 5
Credit: 382
RAC: 0
Sweden
Message 2011142 - Posted: 7 Sep 2019, 21:27:53 UTC

ID: 2011142 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15060
Credit: 4,313,585
RAC: 2,239
Netherlands
Message 2011163 - Posted: 7 Sep 2019, 22:46:36 UTC - in response to Message 2011119.  

You told us the BOINC version. BOINC is the managing program, it'll queue work, upload and download work and does all the scheduling. It doesn't do the calculations, that's done by the project science application.

Normally when you add this project it will determine your operating system and of it has one, send you an appropriate science application. You're running Linux, for which this project has several applications.

However, one can also use their own application or a third party app which is then used in the anonymous platform. Yours is such an app. And so the question is how did you come by it?
ID: 2011163 · Report as offensive
brain

Send message
Joined: 4 Sep 16
Posts: 5
Credit: 382
RAC: 0
Sweden
Message 2011172 - Posted: 7 Sep 2019, 23:47:28 UTC - in response to Message 2011163.  

Included in my list of installed software in my previous post is the seti@home application;

boinc-app-seti (version 8.00~svn4035-1)

It's installed from Debian official repos.
ID: 2011172 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 10998
Credit: 1,108,885,200
RAC: 1,666,661
United States
Message 2011173 - Posted: 8 Sep 2019, 0:00:42 UTC - in response to Message 2011172.  
Last modified: 8 Sep 2019, 0:03:47 UTC

By the version number description, it looks to be pulled from the SVN repository and packaged up by the Debian maintainers.

[Edit] at least for the cpu application. For the client to be at version 7.16.1, either the Costamagna ppa had to be pulled in or the Debian maintainers grabbed the beta client_7.16.1 release at github/boinc and compiled it themselves.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011173 · Report as offensive
brain

Send message
Joined: 4 Sep 16
Posts: 5
Credit: 382
RAC: 0
Sweden
Message 2011182 - Posted: 8 Sep 2019, 1:32:03 UTC - in response to Message 2011173.  
Last modified: 8 Sep 2019, 1:38:47 UTC

By the version number description, it looks to be pulled from the SVN repository and packaged up by the Debian maintainers.


That's how it works...

For the client to be at version 7.16.1, either the Costamagna ppa had to be pulled in or the Debian maintainers grabbed the beta client_7.16.1 release at github/boinc and compiled it themselves.


?

Regarding the boinc client;
7.16.1 was released upstream 31 July and packaged by Debian by 6 August. No betas were involved.
ID: 2011182 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15060
Credit: 4,313,585
RAC: 2,239
Netherlands
Message 2011194 - Posted: 8 Sep 2019, 7:08:23 UTC - in response to Message 2011182.  

A BOINC client release, especially a new major number, is normally released for all platforms. This time only a Linux client was released, because Costamagne built one, all the other platforms are waiting for one addition that needs to be in the client.

As for the Seti app, my advice is to remove it, remove the app_info.xml file, reset the project and run with the project's own default applications first.

If these run better, you have to complain to the Debian maintainer about it.
If the defaults run the same, it's hardware.
ID: 2011194 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 13670
Credit: 183,258,412
RAC: 378,426
United Kingdom
Message 2011195 - Posted: 8 Sep 2019, 8:20:53 UTC - in response to Message 2011182.  

Regarding the boinc client;
7.16.1 was released upstream 31 July and packaged by Debian by 6 August. No betas were involved.
The code for version 7.16.1 was prepared for testing on July 30/31, but not actively released: as Jord says, all platforms are waiting for one addition that needs to be in the client.

Gianfranco and Germano (for LoctutusOfBorg PPA and Fedora respectively) jumped the gun, and built their versions anyway.

Unfortunately, the 31 July code contains a serious bug and should be avoided. The patch for that bug became available on 13 August, and was deployed by Gianfranco and Germano on 16 August: those re-releases (still using the 7.16.1 version number) should be OK for the time being, but will need to be updated in due course.

If the Debian release is dated 6 August, it is likely to contain the 31 July bug, which is dangerous in normal use (all tasks can be aborted whenever the system is shut down). But I don't have direct access to the Debian team.
ID: 2011195 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 10998
Credit: 1,108,885,200
RAC: 1,666,661
United States
Message 2011325 - Posted: 9 Sep 2019, 2:07:00 UTC - in response to Message 2011182.  

That is unusual as the 7.16.1 release is only available as a client release branch still. The master branch is still the 7.14.2 release. The costamagna release has bugs that are being discussed in the issues section of github/boinc. It has unresolved build and dependency issues.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011325 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 13670
Credit: 183,258,412
RAC: 378,426
United Kingdom
Message 2011340 - Posted: 9 Sep 2019, 6:20:50 UTC - in response to Message 2011325.  

The master branch is still the 7.14.2 release.
No. The 'master' branch is the evolving development area, which changes from day to day (and should only be deployed for testing). It still identifies itself as v7.15.0, although that should have been updated to v7.17.0 at the end of July to mark the separation of the v7.16 release and the start of the next cycle of development.

v7.14 is a separate release branch - that branch reached v7.14.2 after a couple of patches.
ID: 2011340 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 10998
Credit: 1,108,885,200
RAC: 1,666,661
United States
Message 2011376 - Posted: 9 Sep 2019, 16:09:03 UTC - in response to Message 2011340.  

The master branch is still the 7.14.2 release.
No. The 'master' branch is the evolving development area, which changes from day to day (and should only be deployed for testing). It still identifies itself as v7.15.0, although that should have been updated to v7.17.0 at the end of July to mark the separation of the v7.16 release and the start of the next cycle of development.

v7.14 is a separate release branch - that branch reached v7.14.2 after a couple of patches.

Duh, of course. Goofed.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011376 · Report as offensive

Questions and Answers : Unix/Linux : Extremely slow computation for some tasks


 
©2020 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.