Message boards : News : Very long jobs
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile Ray Murray
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 281
Credit: 11,859,285
RAC: 2
Message 24713 - Posted: 24 Aug 2012, 21:44:18 UTC - in response to Message 24703.  
Last modified: 24 Aug 2012, 22:27:31 UTC

I need to see Igor about this...........
I set dealine of 30 days always. However I
think BOINC re-computes based on the estimated
FLOPS. Eric.


The deadlines I get are all no more than a week, regardless of estimated completion time, 7hr and 80hr alike. Just now I'm almost exclusively here (with a couple of cores doing occasional T4T for fun) so these deadlines probably aren't a problem for me, unless I get a bunch of long WUs all at once, but most users will probably be sharing with other projects which will lose processor time when Boinc goes into High Priority to squeeze out one of these in time rather than sharing as per the user's preferences, which might upset some. Some Climate models may take 6 months or more but that's what's expected so it isn't a problem and the deadline of up to a year allows for this. The difficulty comes when a 7hr job is expected but a 100hr job arrives.
Think I'll put my slower machine over to T4T once it's finished its 2 long ones just to avoid the deadline issue for now.

From jujube's suggestion: WCG, Climate, YoYo, etc. offer the option to chose different applications of differing lenghts so it might be neccessary to redefine long and short WUs as different applications to allow for the choice.

Happy to see this project properly back into full production again, at last, in time for its 8th birthday. Good to be back crunching again after so many long breaks. Been here almost from the start but I've nearly doubled my score here in the last month alone 8¬)
ID: 24713 · Report as offensive     Reply Quote
Profile Ray Murray
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 281
Credit: 11,859,285
RAC: 2
Message 24714 - Posted: 24 Aug 2012, 21:44:20 UTC - in response to Message 24703.  
Last modified: 24 Aug 2012, 21:48:37 UTC

OOPS fat-finger double post
Please delete this one
ID: 24714 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 24715 - Posted: 25 Aug 2012, 10:16:21 UTC

Great feedback and suggestions. First I am not going to rush
into anything without some planning and review. From CERN
side all is running rather well and I have a lot of work queued.

The next step will be I hope an "LXTRACK" PC for the CERN users.
We are thinking of a second server or project to test NFS comms
between users and BOINC to avoid some AFS issues and avoid
unnecessary file copies. We would also try returning the full results
and not just the minimal "fort.10". This could also be a chance to set up a
new project? or application? LongTrack.

I am just thinking "out loud" here. I particularly appreciate the comparison
with climate prediction. I am definitely out of touch. I shall look for some
relevant papers as I cannot attend BOINC conferences (no travel
budget).

In the meantime back to documentation and result analysis and comparison.

Thanks for your support and comments.

ID: 24715 · Report as offensive     Reply Quote
Profile tazzduke

Send message
Joined: 24 Jun 10
Posts: 34
Credit: 4,962,081
RAC: 10,961
Message 24716 - Posted: 25 Aug 2012, 10:53:11 UTC - in response to Message 24715.  

To Eric and CERN team

All I can say is keep up the good work

Also I got some of those wlxu2 workunits, which for my machine are not a problem, as I crunch one project at a time and considering the workunits available I will be on crunching for awhile :-)

Regards
Tazzduke
ID: 24716 · Report as offensive     Reply Quote
Profile Grubix

Send message
Joined: 3 Jul 08
Posts: 20
Credit: 8,281,604
RAC: 0
Message 24717 - Posted: 25 Aug 2012, 13:01:21 UTC - in response to Message 24716.  

Here's my experience on the topic very long WUs, maybe that's a suggestion for LHC@Home: There are users who are looking forward to a long WUs because they have new and very fast hardware. On my i7-3930K I'd love to calculate a few of the long WUs. In some projects, there is even such a thing as competition, who has the longest.

Then there are users with very slow hardware. I have logged out my Athlon 1800 from LHC, because he can not do the long WUs in the deadline. These users would be happy to get short WUs.

Like the other projects have solved? In the Preferences there are three optional settings:

Run short tasks: yes / no
Run long tasks: yes / no
If no work for selected applications is available, accept work from other applications: yes / no

With these options, you can then decide for yourself what you want. For the project it has several advantages:

- Users don't complain about the long WUs.
- Even older hardware can be running, the project can use more computers.
- Mainly users with faster hardware will select long WUs. Therefore you get the long WUs back faster and not have to wait maybe 30 days on a P4.

Some projects also have the option "Run beta tasks: yes / no". These tasks you get only when you have turned on the option (even though "If no work for selected applications is available ..." is set to yes). Here, the project can sometimes try something more unusual things. *lol*

The default settings are often:

Run short tasks: yes
Run long tasks: no
Run beta tasks: no
If no work for selected applications is available, accept work from other applications: yes

How this is done at the server, I do not know. But maybe e.g. GPUGRID can help.

Just an idea from me, Grubix.
ID: 24717 · Report as offensive     Reply Quote
Sunny129
Avatar

Send message
Joined: 12 Dec 05
Posts: 31
Credit: 9,709,398
RAC: 0
Message 24718 - Posted: 25 Aug 2012, 14:21:27 UTC

the two machines i have on LHC@H right now are 6-core Thuban chips, so finishing these long WUs is never a problem, even if the deadline is no longer than ~7 days from the initial DL. however, b/c the deadlines for the long WUs are too short, BOINC is kicking them into high priority mode, which IS causing problems for SETI@Home and the Wow! Event that i'm participating in right now...specifically, my dual GTX 560 Ti host won't DL any new SETI@Home work b/c LHC@H is running in high priority mode.

i think jujube, Ray, and Grubix have mentioned the best possible solution - give us the choice to run either short WUs, long WUs, or both types of WUs...and an option to run other applications when the selected ones aren't available. of course this won't really help me w/ my SETI@Home issues, since the Wow! Event ends on the 29th, and such a server implementation would no doubt take longer than that. also, i don't know if these long WUs are going to be a regular thing, or if its just a one time thing, in which case it might not be worth it to make those changes to the project server...
ID: 24718 · Report as offensive     Reply Quote
Toby Broom
Volunteer moderator

Send message
Joined: 27 Sep 08
Posts: 798
Credit: 642,809,001
RAC: 284,014
Message 24720 - Posted: 25 Aug 2012, 16:53:46 UTC

I like the idea of putting options in the server and keepting things under this BOINC project.

Great work to iron out the kinks and get this project back on it's feet.

Also, great communication with the comunity!
ID: 24720 · Report as offensive     Reply Quote
Christoph

Send message
Joined: 25 Aug 05
Posts: 69
Credit: 306,627
RAC: 0
Message 24723 - Posted: 26 Aug 2012, 9:55:01 UTC
Last modified: 26 Aug 2012, 9:55:26 UTC

One thing other projects do is to bundle up to 8 tasks into one to produce long running WUs.

The long ones are thought especially for GPUs since they cut down the processing time a lot.

Reference for that is Collatz@home. They have 'mini' collatz. That is just one WU.
Christoph
ID: 24723 · Report as offensive     Reply Quote
[AF>FAH-Addict.net]toTOW

Send message
Joined: 9 Oct 10
Posts: 77
Credit: 3,671,357
RAC: 0
Message 24724 - Posted: 26 Aug 2012, 10:31:04 UTC - in response to Message 24715.  

In my opinion, it would be better to add these experiments as new applications to the existing project. I don't think multiplying independent projects is a good idea.
ID: 24724 · Report as offensive     Reply Quote
Profile Tom95134

Send message
Joined: 4 May 07
Posts: 250
Credit: 826,541
RAC: 0
Message 24809 - Posted: 9 Sep 2012, 19:49:51 UTC

Back on-board after a week long holiday on the north coast of California in the fog and clouds.

I got four 80 hours duration Tasks with a Deadline of 9/15. These are running at Normal Priority (1 is waiting to run). However, four other Tasks (short durations of about 4 hours) are running at High Priority. This is one I don't understand because it was my understanding that BOINC was smart enough to figure out which Tasks needed to be pushed to High Priority.

Also running T4T (one Task) and SETI (GPU) which both appear to be running at Normal priority.

Windows 7, Intel i7-2600, 16GB RAM, NVIDIA

Tom
ID: 24809 · Report as offensive     Reply Quote
Profile Tom95134

Send message
Joined: 4 May 07
Posts: 250
Credit: 826,541
RAC: 0
Message 24810 - Posted: 10 Sep 2012, 1:48:45 UTC

Short Tasks have finished and now the long ones are at High Priority/
ID: 24810 · Report as offensive     Reply Quote
lugal

Send message
Joined: 8 Aug 08
Posts: 6
Credit: 239,602
RAC: 0
Message 24823 - Posted: 13 Sep 2012, 8:26:08 UTC

I received a long wu yesterday (approx. 60 hours runtime) but the deadline is September 16th.

This is not feasible as my machine can do jobs only max. 12 hours a day.

Do you propose that I abort the wu?

It passed through several hands already.

Regards lugal
ID: 24823 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 24824 - Posted: 13 Sep 2012, 9:22:32 UTC - in response to Message 24823.  

Please feel free to abort it. I am not submitting any
more "very long" for the moment. Nonethless they
have run rather successfully and I have received a
tremendous amount of feedback. Will post news when I
have a solution. Eric.
ID: 24824 · Report as offensive     Reply Quote
Profile Tom95134

Send message
Joined: 4 May 07
Posts: 250
Credit: 826,541
RAC: 0
Message 24829 - Posted: 13 Sep 2012, 17:45:34 UTC

I have three Tasks that are going to run 50+ hours (Deadline 9/15) that will finish today. Note: these started out indicating a Remaining time of 80 hrs.

Tom
ID: 24829 · Report as offensive     Reply Quote
Werkstatt

Send message
Joined: 5 Oct 08
Posts: 12
Credit: 1,108,455
RAC: 0
Message 24831 - Posted: 13 Sep 2012, 18:12:51 UTC
Last modified: 13 Sep 2012, 18:13:57 UTC

what could be done is:
give the volunteers the choice to select which apps should be cruched. Seti does that and it works fine
https://dl.dropbox.com/u/50246791/selected%20apps.PNG
ID: 24831 · Report as offensive     Reply Quote
Previous · 1 · 2

Message boards : News : Very long jobs


©2024 CERN