climateprediction.net home page
A task just timed out, and it was mostly the scheduler on my machine, I think.

A task just timed out, and it was mostly the scheduler on my machine, I think.

Message boards : Number crunching : A task just timed out, and it was mostly the scheduler on my machine, I think.
Message board moderation

To post messages, you must log in.

AuthorMessage
Jean-David Beyer

Send message
Joined: 5 Aug 04
Posts: 1120
Credit: 17,198,774
RAC: 2,444
Message 42722 - Posted: 30 Jul 2011, 20:40:32 UTC

I received task ID 12846641 on 30 Apr 2011 12:32:00 UTC. As I recall it required 2000 to 3000 hours estimated run time and I figured out it could probably just manage it with a little time to spare. I usually run this machine 24/7. It has 2 3.06 GHz Hyperthreaded Xeon processors (the ones with a 1 GByte L3 Cache). The machine has 8 GBytes RAM. Climate Prediction is set to get 47% of the time boinc client receives. It has returned lots of trickles.

Now the Boinc Client, I think it is called, is set up to run 4 processes at a time, and lately the machine has not been doing much except web browsing and e-mail. In any case, the client gave very little time to this task for about a month, fetching new tasks and running them in preference. These were short tasks (a few hours heach). But it kept ignoring this task for another week or so and then set it to high priority and gave it full time, but by then it was too late. I have let it run. It expired today and if I look at the task on this web site, it says, among other things, Timed out - no response. However, it is still running on my machine. Should I let it continue, or should I cancel it? I am disappointed thatBoinc client 6.10.58 did not give it full time all the time.
ID: 42722 · Report as offensive     Reply Quote
Profile Greg van Paassen

Send message
Joined: 17 Nov 07
Posts: 142
Credit: 4,271,370
RAC: 0
Message 42723 - Posted: 30 Jul 2011, 21:09:00 UTC - in response to Message 42722.  

Please keep running it, Jean-David. Scientists will use the result.
ID: 42723 · Report as offensive     Reply Quote
Jean-David Beyer

Send message
Joined: 5 Aug 04
Posts: 1120
Credit: 17,198,774
RAC: 2,444
Message 42724 - Posted: 31 Jul 2011, 0:46:27 UTC - in response to Message 42723.  

OK, I will let it finish.

Is there any explanation for why the client seems to have silently refused to run the application when there would have been CPU time to run it? It would continually download other applications from other projects that kept crowding climate prediction away from the processors. I suppose it is a boinc client problem, not really a climate prediction one.

I also have another climate prediction application that is not likely to finish on time. I will let it run too. It is this one:

13123966 7557477 845454
UK Met Office Coupled Model Full Resolution Ocean v6.07

It has a little over 2411 hours to go. It is running at high priority. It is about 25% done. But this one may have been too long from the beginning. I was not paying much attention to it.

Last time contacted server 30 Jul 2011 19:10:33 UTC
% of time BOINC client is running 95.1179 %
While BOINC running, % of time work is allowed 99.9929 %
Task duration correction factor 1.243636
ID: 42724 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 42725 - Posted: 31 Jul 2011, 1:19:08 UTC - in response to Message 42724.  

I am disappointed that Boinc client 6.10.58 did not give it full time all the time.

That's probably because of the way that you've set your preferences:
Climate Prediction is set to get 47% of the time boinc client receives


Time share allocation may account for all of the behaviour that you've seen.
To "be fair" to all of your projects, the time should be 100% for each project.

As for the messages that end up on the server page for models, it uses the first one that the server gets from a computer.
Most projects don't allow work units to go past deadlines, so the software hasn't been designed to allow the message to be updated by later messages.
This project is probably unique in the way that it works, with a lot of the BOINC code being ignored, in favour of custom written code. (cpdn existed long before BOINC came out.)


Backups: Here
ID: 42725 · Report as offensive     Reply Quote
Jean-David Beyer

Send message
Joined: 5 Aug 04
Posts: 1120
Credit: 17,198,774
RAC: 2,444
Message 42727 - Posted: 31 Jul 2011, 11:29:09 UTC - in response to Message 42725.  

I run 5 projects. I want climate prediction to get half the time and the other four to get 1/8 of the time each. Since I have four processors (or 2, depending on how you count hyperthreaded processors), this should allow two climate prediction processes to run and two others all the time, statistically speaking. And this is usually what happens. But I have noticed that the very long climate prediction ones seem to get deferred a lot until they seriously risk timing out. I think it is only climate prediction ones that have estimated run times 2000 hours or more. One of these has a deadline in late 2012 and should have no difficulty in achieving this. It is 31% complete and has only about 2400 hours to go. But others have deadlines only a little later than the time they need if run all the time. And they would run if the Boinc Client did not download so many short tasks with very close deadlines.

But sometimes it will download 10 or more malaria control or rosetta tasks. These normally take only 3 hours each or so to complete, and have short (2 or 3 day) deadlines, but if it continually downloads these, it postpones the climate prediction ones. Pretty soon the climate prediction ones go into high priority mode, but sometimes if the short ones get too close to deadline, they go into high priority mode, climate prediction gets into normal mode and wait.

My settings are made so they all total to 100%. I.e.,

climate prediction 47%
malaria control 17%
rosetta 16%
Seti@home 10%
World Comm.Grid 10%

The numbers are not 50, 12.5, 12.5, 12.5, 12.5 because they accumulate credits at different rates, and these settings tend to give me better results.
ID: 42727 · Report as offensive     Reply Quote

Message boards : Number crunching : A task just timed out, and it was mostly the scheduler on my machine, I think.

©2024 cpdn.org