What should I do if a task goes past the deadline?

Message boards : BOINC Manager : What should I do if a task goes past the deadline?
Message board moderation

To post messages, you must log in.

AuthorMessage
arakelov

Send message
Joined: 11 Jul 21
Posts: 6
Spain
Message 104806 - Posted: 16 Jul 2021, 13:52:07 UTC

Hi! I've been using BOINC for a couple of months and I still have some doubts about the timings of the tasks, for example, look at those selected:



According to the estimated remaining time, these tasks will long pass the deadline (tomorrow). In that case, what will happen? Will these tasks be immediately assigned to someone else after the deadline? Should I let them finish? If so, will the server reject them when it receives them? Have I lost 4 days of computation?

I can't understand why the estimated times don't match the corresponding deadlines, my CPU+GPU are running nearly 100% most time.

Thanks a lot in advance!
ID: 104806 · Report as offensive
Profile Dave
Help desk expert

Send message
Joined: 28 Jun 10
Posts: 2563
United Kingdom
Message 104820 - Posted: 17 Jul 2021, 13:58:59 UTC - in response to Message 104806.  

According to the estimated remaining time, these tasks will long pass the deadline (tomorrow). In that case, what will happen? Will these tasks be immediately assigned to someone else after the deadline? Should I let them finish? If so, will the server reject them when it receives them? Have I lost 4 days of computation?


It depends on the individual project. Some you won't get the credit for and some you will. CPDN for example you still get the credit long after the results are any real use. (The researcher wanting the results will have finished their PhD that would have used the results long before some results come in.)

For definitive answers you are best asking with the individual projects.

Sometimes computers can report their speed incorrectly resulting in tasks being sent that will not complete in time or of course the amount of time the computer is running or other work on the computer can also cause this.
ID: 104820 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 104821 - Posted: 17 Jul 2021, 16:07:59 UTC
Last modified: 17 Jul 2021, 16:08:38 UTC

I wouldn't run PrimeGrid from CPU.
It runs too slow.
It's a project most people run on a GPU.
A modern GPU will finish each WU in minutes to just a few hours tops, and far exceeds any desktop CPU in performance, including Ryzen Threadrippers, for a lot less power.
My recommendation is to run it via GPU, and avail your CPU threads for projects that have no GPU alternative..
ID: 104821 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5082
United Kingdom
Message 104823 - Posted: 17 Jul 2021, 16:27:39 UTC

The first, and most important change, is to reduce the amount of work you are requesting - so it doesn't happen again.

Then, consider your project settings. Your screenshot says you are running 10 'mt' (multi-threaded) tasks at the same time. Normally, an mt task - by itself - uses all CPU cores available in the machine, so that would be extremely inefficient. However, PrimeGrid is a special case, and you can set separate core limits for each search type.

Count up your core demand and allocation carefully, and make sure you aren't - literally - overcooking your CPU.

There is a particular problem when mt tasks are allocated by the server, which I've written up at GitHub #4151 - using PrimeGrid as one of the examples. When calculating how much work to send in response to your work request, the server doesn't take any account of the number of cores the task(s) will occupy - so it tends to send 4, 8, 12 ... times as much work as you were asking for, depending on the value of your 'cores per task' setting.

So - back to the first point - ask for a small amount of work overall, and small amounts at each subsequent request.
ID: 104823 · Report as offensive
arakelov

Send message
Joined: 11 Jul 21
Posts: 6
Spain
Message 104825 - Posted: 17 Jul 2021, 20:37:00 UTC - in response to Message 104823.  

Thanks to everyone for your advice!

@ProDigit: I'm just running two projects: PrimeGrid and NFS@Home. Following your indications, I'm changing PrimeGrid to run only GPU tasks and letting CPU for NFS@Home tasks. (Is this a good choice?)
@RichardHaselgrove: In fact I often see up to 16 tasks with status='Running', I think 1 for each CPU thread (htop confirms that). Yes, my CPU is quite hot, sometimes over 90ºC! Fortunalely, I have a good cooling. I'm very newbie to BOINC and those projects, so I would ask for a page where I can learn to configure BOINC properly, specially these parameters about "amount of work I'm requesting". Does it have to do with BOINCMgr > Options > Computing preferences > Computing > Use at most x% of the CPUS / CPU time? Or is it project-dependant?

Thanks again, regards.
ID: 104825 · Report as offensive
JT

Send message
Joined: 15 May 20
Posts: 8
Message 104826 - Posted: 17 Jul 2021, 20:50:10 UTC - in response to Message 104821.  

I wouldn't run PrimeGrid from CPU.
It runs too slow.
It's a project most people run on a GPU.
A modern GPU will finish each WU in minutes to just a few hours tops, and far exceeds any desktop CPU in performance, including Ryzen Threadrippers, for a lot less power.
My recommendation is to run it via GPU, and avail your CPU threads for projects that have no GPU alternative..

Quoted for the LOL !
ID: 104826 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 104987 - Posted: 8 Aug 2021, 0:15:50 UTC - in response to Message 104825.  

Thanks to everyone for your advice!

@ProDigit: I'm just running two projects: PrimeGrid and NFS@Home. Following your indications, I'm changing PrimeGrid to run only GPU tasks and letting CPU for NFS@Home tasks. (Is this a good choice?)

Thanks again, regards.

I would guess, since NFS is CPU only, it's your best configuration.
Prime grid might get you more PPD than NFS (or not, not sure), but even if, you'll be able to get a more even load distribution.
Prime Grid WUs done on CPU are only a fraction of what can be done on a GPU.
ID: 104987 · Report as offensive
KeroseneBlast

Send message
Joined: 6 Oct 21
Posts: 2
Message 105677 - Posted: 6 Oct 2021, 21:38:17 UTC

Is there a way to change what task is running? I have a mt task that's not due for a few weeks, but BOINC is running it instead of smaller tasks that are due sooner.
ID: 105677 · Report as offensive
Les Bayliss
Help desk expert

Send message
Joined: 25 Nov 05
Posts: 1654
Australia
Message 105678 - Posted: 6 Oct 2021, 21:48:19 UTC - in response to Message 105677.  

BOINC is designed to juggle a lot of parameters, and the case you mention is one of them.
If another task gets close to running out of time, BOINC will switch to it.

You can fiddle with things yourself, but then you'll need to watch what's happening all the time.

Also, don't try caching more work than the processors can handle. That's just asking for trouble.
Take lots of small bites, not one big one. :)
ID: 105678 · Report as offensive
KeroseneBlast

Send message
Joined: 6 Oct 21
Posts: 2
Message 105679 - Posted: 7 Oct 2021, 0:20:17 UTC - in response to Message 105678.  

You'd think it would, but it's already blown past a deadline that it could've hit if it switched and still hit the other target's deadline. Is there a way to manually switch tasks?
ID: 105679 · Report as offensive
Les Bayliss
Help desk expert

Send message
Joined: 25 Nov 05
Posts: 1654
Australia
Message 105680 - Posted: 7 Oct 2021, 0:49:33 UTC - in response to Message 105679.  

In the Tasks tab, Suspend the one that you want to stop.
Let the others run.
Unsuspend the task that you stopped earlier.
ID: 105680 · Report as offensive
Bryn Mawr
Help desk expert

Send message
Joined: 31 Dec 18
Posts: 287
United Kingdom
Message 105711 - Posted: 8 Oct 2021, 6:13:35 UTC - in response to Message 104825.  

Thanks to everyone for your advice!

@ProDigit: I'm just running two projects: PrimeGrid and NFS@Home. Following your indications, I'm changing PrimeGrid to run only GPU tasks and letting CPU for NFS@Home tasks. (Is this a good choice?)
@RichardHaselgrove: In fact I often see up to 16 tasks with status='Running', I think 1 for each CPU thread (htop confirms that). Yes, my CPU is quite hot, sometimes over 90ºC! Fortunalely, I have a good cooling. I'm very newbie to BOINC and those projects, so I would ask for a page where I can learn to configure BOINC properly, specially these parameters about "amount of work I'm requesting". Does it have to do with BOINCMgr > Options > Computing preferences > Computing > Use at most x% of the CPUS / CPU time? Or is it project-dependant?

Thanks again, regards.


It’s BOINCMgr > Options > Computing preferences > Computing > Store at least

and

BOINCMgr > Options > Computing preferences > Computing > Store up to an additional

Set both parameters to, maybe, 0.1 and work from there.
ID: 105711 · Report as offensive

Message boards : BOINC Manager : What should I do if a task goes past the deadline?

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.