climateprediction.net home page
infinite Work Unit

infinite Work Unit

Questions and Answers : Windows : infinite Work Unit
Message board moderation

To post messages, you must log in.

AuthorMessage
dougga

Send message
Joined: 27 Nov 06
Posts: 5
Credit: 310,754
RAC: 0
Message 30674 - Posted: 24 Sep 2007, 6:18:57 UTC

I have a dual core intel T2500 and have two climateprediction work units going. They appear to be never-ending bits of work with the time to completion ticking up at the same rate that the cpu time used.

WU #1: cpu time: 501:24:45 [increasing]
progress: 2.381%
to completion: 860:31:19 [increasing]

This suggest just 20 days of work which is ridicluous. This thing has been going for months.

Can anyone help me with this?
Is this a bogus work unit?
Is my client broken?
Did the oopma loopa\'s swap out my WUs for the everlasting gobstoppers?
ID: 30674 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 30675 - Posted: 24 Sep 2007, 6:57:21 UTC


Well, it\'s a slow processor, and it\'s a laptop, so I\'m guessing that it doesn\'t get run for long periods.

The speed of crunching depends on:
1) the processor speed,
2) how long each day it\'s left on,
3) how much of this time is used by YOU, (because the model only runs at low priority, using up spare processor time),
4) if you have the General preferences option: \"Leave applications in memory while suspended? \" set to yes, (to prevent loss of crunch time when all of the memory is needed by another program),
5) and if you wait until just after the data is saved before shutting down the computer.
(In the Show graphics display, there is a count down timer, which can be seen by pressing: Z then 8. Wait until the timer reaches zero and then returns to 432 and starts count again.)

Finally, the BOINC estimates are only accurate for the work units of shorter projects, which last a few hours to a few days. For climate models, they are way off, and only start to get realistic after a few models have been completed, so that BOINC KNOWS how long it takes, instead of having to rely on estimates.

So, just keep plodding away. You\'ll get there eventually.


Backups: Here
ID: 30675 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 30676 - Posted: 24 Sep 2007, 7:41:09 UTC
Last modified: 24 Sep 2007, 7:44:15 UTC


I usually ignore the \'to completion\' column and work it out using the %progress and CPU time so far. In this case, if the numbers are correct, you\'re about 1/40th of the way through with 500 hours, implying a total of around 20,000 hours.

As Les says, if the model is being taken out of memory before it writes it, progress since the last checkpoint will be lost.

Another possible cause for having it run slowly is \'thermal throttling\', which is when the CPU deliberately slows itself down when it is about to overheat. Some laptops don\'t have very good cooling, and as a result aren\'t really suitable for Boinc.

The trickle list for the most advanced model is here, and shows a s/ts time of 60 seconds on your slab model (how many timesteps per checkpoint are there for slab?).

http://climateapps2.oucs.ox.ac.uk/cpdnboinc/result.php?resultid=6631684

I have several suggestions:
* It might be worthwhile downloading and running Prime95\'s Torture Test on the PC to make sure that it is running OK. You need to run it for 24 hours, with a copy for each CPU core.
* Try running just one climate model at a time on that PC (click \'no more work\' and abort the one with least CPU time).
* If these ideas don\'t help, it might be better to use this PC for shorter duration projects.

I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 30676 · Report as offensive     Reply Quote

Questions and Answers : Windows : infinite Work Unit

©2024 cpdn.org