climateprediction.net home page
BBC Climate prediction takes along time!

BBC Climate prediction takes along time!

Questions and Answers : Unix/Linux : BBC Climate prediction takes along time!
Message board moderation

To post messages, you must log in.

AuthorMessage
John Price
Avatar

Send message
Joined: 6 Aug 04
Posts: 66
Credit: 7,038,260
RAC: 11,435
Message 21595 - Posted: 25 Mar 2006, 20:34:14 UTC

I have two of these running on Linux boxes now and one of them is showing it will take 4,648 hours to complete with a deadline of 26 Feb 2007.
After 78 hours it is showing completed 1.73%.
My concern is that as I don\'t run this box 24/7 I think the likelihood of me completing this is remote. I don\'t want to run this box just to complete clim pred project as that seems to be just adding to global warming.
Is there any point in continuing this in as much as it is unlikly it will be completed in the alloted time?
Computer 314638
model hadcm3lb 5.08
BOINC 5.2.13
AMD 200+
512mg RAM
ID: 21595 · Report as offensive     Reply Quote
Profile old_user248

Send message
Joined: 6 Aug 04
Posts: 65
Credit: 1,605,224
RAC: 0
Message 21602 - Posted: 25 Mar 2006, 23:20:16 UTC

The results will be used anyway. As far as I know that date is for priority setting, but I am probably wrong and will get pounced on because of it :)

I generally just let my Linux boxes run.

Dave
ID: 21602 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 21607 - Posted: 26 Mar 2006, 2:39:26 UTC

You\'re close Dave.
The deadline is included because it\'s a necessary part of other projects, so something had to be put into that spot in the software.

John
Every 10 years completed is valuable. And if you can get to the half way point, that\'ll mean that you\'ve completed the hindcast part. Of even more value.

And BOINC V5.* isn\'t very good at estimating completion times for very long projects, because the method was changed to \'do averages\' over several models. Not a good method with climate models.
Try DIY, using the real world time that it takes to complete 1 model year, then multiply by 160. A better estimate will be to use 10 model years when you get that far.

ID: 21607 · Report as offensive     Reply Quote
A.T.Hofkamp

Send message
Joined: 25 Nov 05
Posts: 1
Credit: 1,372,651
RAC: 0
Message 22728 - Posted: 12 May 2006, 11:23:01 UTC

I also found that the estimation is way off, mine is at 4495 hours at the moment.

Instead I use the percentage and the used CPU time to compute the estimated #hours as follows:
456.5 hours, 13.46% currently done for my model means that total computation time will be 456.5*100/13.46 = 3391.5 hours CPU time.
This number is fairly constant, it is always around 3300 hours.

So after 3391.5-456.5 CPU hours = 2935 CPU hours = 122 CPU days the model will have finished.

Since I run climateprediction for 80% of the time that is
122/0.8 = 152 real days = about 5 months.

Albert
ID: 22728 · Report as offensive     Reply Quote

Questions and Answers : Unix/Linux : BBC Climate prediction takes along time!

©2024 cpdn.org