climateprediction.net home page
HadCM3 6.04 - 9,000 hour job?

HadCM3 6.04 - 9,000 hour job?

Questions and Answers : Windows : HadCM3 6.04 - 9,000 hour job?
Message board moderation

To post messages, you must log in.

AuthorMessage
Adrian Booth

Send message
Joined: 10 Sep 08
Posts: 6
Credit: 2,587,542
RAC: 0
Message 36287 - Posted: 4 Mar 2009, 8:28:36 UTC

Hi

I just received a job

HadCM3 Couple Model Experiment Optimised File I/O 6.04

hadcm3istd_cq3q_1920_160_06017962_2

which, while it has not started running yet, is estimated to require 9060 hours to complete (just over a year of 24x7 crunching).

I don\'t have a problem with this; but is it normal?

Cheers
Adrian
ID: 36287 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 36288 - Posted: 4 Mar 2009, 8:42:29 UTC

The other 2 that you\'re running are both shorter. This may have messed up BOINC\'s estimate of the time to complete. When the hadcm3 starts, the estimate should start dropping, although this may take 30% or more of running to start showing up.

It\'s always rather tricky to get a close estimate at the start. The more models of a given type that BOINC runs, the better it will get at an accurate estimate.

(A 160 year model took 4 months on my old 3.2GHz P4, running by itself.)


Backups: Here
ID: 36288 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 36289 - Posted: 4 Mar 2009, 9:04:46 UTC
Last modified: 4 Mar 2009, 9:06:00 UTC

I just thought of something else: BOINC may be including time for running 3 models on 2 processors, with models alternating in their processing.

In case you\'re not aware: With these long models, it\'s usually a good idea to keep the project set for No new tasks once one model per processor has been downloaded. This can be turned back on for long enough for a new model to be requested when a current model has finished.
And, now that you do have 3, you can Suspend the long model until one of the others finishes. AFTER you\'ve set No new tasks!

NOTE: Allow new tasks only has to be there long enough for BOINC to REQUEST a new model, not for all of the time needed to DOWNLOAD it.
Backups: Here
ID: 36289 · Report as offensive     Reply Quote
wateroakley

Send message
Joined: 6 Aug 04
Posts: 185
Credit: 27,083,655
RAC: 6,161
Message 36290 - Posted: 4 Mar 2009, 9:08:00 UTC

Adrian, if this is the model 8377076 running on Q6400 910240, the PC has already run a similar 160 year CM3 model 8099111 from 26 Sep 2008 to 7 Feb 2009, at 2.37 s/TS, using 2729 cpu hours.

The new one is another long 160 year model, your last one took about 4.3 months, so the 9060 hours is perhaps misleading. If the PC is on 24/7 at 2.37 s/TS it should be around 3.7 months.
HTH.
ID: 36290 · Report as offensive     Reply Quote
Adrian Booth

Send message
Joined: 10 Sep 08
Posts: 6
Credit: 2,587,542
RAC: 0
Message 36292 - Posted: 4 Mar 2009, 10:54:48 UTC - in response to Message 36290.  

Thanks Les and Hagar. I artificially downloaded a model a bit early (40 odd hours to go on one job) to make sure I didn\'t run out of work on this PC, as it has its Internet access blocked most of the time (my son finds the Internet very distracting from his studies).

What difference does disabling new tasks make? I normally have a 3 day work buffer and only run climateprediction.net on this PC. I would have thought BOINC would not do anything until I had less than 3 CPU days work remaining?

Cheers
Adrian
ID: 36292 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 36295 - Posted: 4 Mar 2009, 19:40:09 UTC

Disabling enables you to suspend one or more tasks without BOINC downloading more.

I mentioned this in case all three of your current models were being processed alternately. This would allow you to complete the 2 that have been running for a while, and THEN starting the long model.


Backups: Here
ID: 36295 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 36296 - Posted: 4 Mar 2009, 19:56:29 UTC
Last modified: 4 Mar 2009, 19:57:16 UTC

Adrian said

\'What difference does disabling new tasks make? I normally have a 3 day work buffer and only run climateprediction.net on this PC. I would have thought BOINC would not do anything until I had less than 3 CPU days work remaining?\'

That should always be the case, but the most recent versions of BOINC (6.4.* and 6.6.*) have had serious work fetch problems in alpha testing and to tell you the truth I wouldn\'t yet trust them always to fetch the right amount of work even in their public releases. I/we (meaning the CPDN mods) may of course be unduly suspicious!
Cpdn news
ID: 36296 · Report as offensive     Reply Quote
Adrian Booth

Send message
Joined: 10 Sep 08
Posts: 6
Credit: 2,587,542
RAC: 0
Message 36303 - Posted: 5 Mar 2009, 2:21:09 UTC - in response to Message 36295.  

Disabling enables you to suspend one or more tasks without BOINC downloading more.

I mentioned this in case all three of your current models were being processed alternately. This would allow you to complete the 2 that have been running for a while, and THEN starting the long model.



Thanks Les - that makes sense. Appreciate your feedback.

Cheers
Adrian
ID: 36303 · Report as offensive     Reply Quote
Profile JIM

Send message
Joined: 31 Dec 07
Posts: 1152
Credit: 22,053,321
RAC: 4,417
Message 36318 - Posted: 6 Mar 2009, 6:34:47 UTC

Another reason for clicking “No New Tasks” is that if the model should crash not allowing new tasks prevents the manager from downloading a replacement before you have time to restore from a backup. It’s frustrating to find a new WU already downloaded and running when you have a perfectly good backup of the old WU.

ID: 36318 · Report as offensive     Reply Quote
old_user92639

Send message
Joined: 13 Aug 05
Posts: 54
Credit: 117,227
RAC: 0
Message 36328 - Posted: 7 Mar 2009, 11:28:53 UTC - in response to Message 36290.  

Adrian, if this is the model 8377076 running on Q6400 910240, the PC has already run a similar 160 year CM3 model 8099111 from 26 Sep 2008 to 7 Feb 2009, at 2.37 s/TS, using 2729 cpu hours.

The new one is another long 160 year model, your last one took about 4.3 months, so the 9060 hours is perhaps misleading. If the PC is on 24/7 at 2.37 s/TS it should be around 3.7 months.
HTH.



Q6400 910240 <= Cache: 488.28 KB = 50%, 1 MB = 100%

:p
ID: 36328 · Report as offensive     Reply Quote

Questions and Answers : Windows : HadCM3 6.04 - 9,000 hour job?

©2024 climateprediction.net