climateprediction.net home page
HADCM3 6.04 "to Completion time" increases each day?

HADCM3 6.04 "to Completion time" increases each day?

Questions and Answers : Windows : HADCM3 6.04 "to Completion time" increases each day?
Message board moderation

To post messages, you must log in.

AuthorMessage
John Mead

Send message
Joined: 14 Jul 07
Posts: 1
Credit: 6,101,468
RAC: 296
Message 37420 - Posted: 4 Jul 2009, 17:07:22 UTC

I have noticed the last few days that the "percent completed" increases, the elapsed time increases, but the time "to completion" actually fluctuates with a trend moving upwards/increasing. I am running win vista 64 on a Athlon dual core. The computer is running at 100% - both cores. I have each core running a separate data set.

This seems pretty weird to me. Especially after several days of this. I can believe it would change some. This is a glass floor I cannot get through after several days at 100% both CPU's.
ID: 37420 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,403,322
RAC: 5,085
Message 37421 - Posted: 4 Jul 2009, 18:08:39 UTC

Your models are running at about 2.7 s/timestep. There are 4147200 timesteps in a 160 year hadcm3 model. That works out to about 3100 hours total for a model to complete. The initial estimate for a model completion by BOINC is often very wrong, it therefore adjusts it as more time is spent on the model. For a reasonably accurate estimate on how much longer it would be to complete it, just subtract your current CPU time from 3100 hours.
ID: 37421 · Report as offensive     Reply Quote
old_user122149

Send message
Joined: 30 Nov 05
Posts: 7
Credit: 42,511
RAC: 0
Message 38924 - Posted: 17 Feb 2010, 10:22:29 UTC - in response to Message 37421.  

can the initial estimate be fixed a little or the running to completion be corrected quicker please?

So the only accurate measurement is the time elapsed? (not the percent complete or time till completion)


First time I have started a CP workunit for some time now and I had to come here to find out what the issue was.
ID: 38924 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 38925 - Posted: 17 Feb 2010, 11:11:16 UTC

This thread is about the long (80/160 year) models. These are no longer available.

The ones that you\'re crunching only take about 4 days, and the time to completion is fairly accurate.


Backups: Here
ID: 38925 · Report as offensive     Reply Quote
old_user122149

Send message
Joined: 30 Nov 05
Posts: 7
Credit: 42,511
RAC: 0
Message 39392 - Posted: 27 Mar 2010, 15:11:24 UTC - in response to Message 38925.  
Last modified: 27 Mar 2010, 15:21:33 UTC

This thread is about the long (80/160 year) models. These are no longer available.

The ones that you\'re crunching only take about 4 days, and the time to completion is fairly accurate.


Yah, not true - it said 100 odd hours to start with and with a deadline of 1 year in the future.
Nearly completed it - but at 336 minutes so far I think you\'ll agree that the estimate was wildly off, coupled with the non informatory (being generous here) deadline, this \'estimation\' of time involved is not FAIRLY ACCURATE.

The factor of inaccuracy is ~3.3.
This crap exceeds my tolerance levels and once this wu gets done i will downrate this project.
I guess you can state categorically that the CP WU completion estimates actually use the hosts benchmark resutls? If so - it needs correcting. if not then to call it an estimate is bordering on the deceptive.

Plus this boinc client is straight taking the piss:

I got to the last \'26mins\' of the CP wU and boinc decides to download a WU from another project - fine - howerver it then runs the WU.


Furthermore after that one is complete is downloads an additional WU ..and starts running that.


(and it does not even bother clearing the completed wu from the screen)

Good day.
ID: 39392 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 39395 - Posted: 27 Mar 2010, 15:37:22 UTC

If that is your experience with the models, then there\'s something wrong with the settings that you\'re using.
Perhaps you have a very low setting for Resource share, for instance.

At 14.7213, your s/TS (seconds per timestep) is VERY slow.

My models of that type were running at around this speed: 3.1580 s/TS
This was on an Intel Q6600 at stock speed, (2.40GHz), running 24/7, with no other projects.


Backups: Here
ID: 39395 · 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 39396 - Posted: 27 Mar 2010, 15:42:34 UTC
Last modified: 27 Mar 2010, 15:42:54 UTC

Hi Behemoth

It hadn\'t taken 336 minutes so far. It looked as if it had taken 336 hours. But that\'s elapsed time. If you highlight the task and click the Properties button you see its CPU time which is the actual processing time. The CPU time is also displayed on the model\'s web page.

The 1,059,930sec displayed at or near the end means 295 hours. Yes, this is much longer than the initial estimate. Looking at the web page again I see that on your computer this model processed very slowly, at 14+ sec per timestep. I don\'t know why your computer was so slow but I\'m sure this explains the discrepancy.

After a model has completed it remains in the Boinc manager saying Ready to report in the Status column. Within 24 hours it should report to the server and disappear from your manager.

Do please downrate ClimatePrediction if it\'s getting on your nerves, though of course we will miss you.
Cpdn news
ID: 39396 · Report as offensive     Reply Quote

Questions and Answers : Windows : HADCM3 6.04 "to Completion time" increases each day?

©2024 climateprediction.net