Hey,

I heard someone saying he was getting way better results using one of the applications compared to the other ones.

Is there any rule as to what application uses what type of configuration most efficiently?

This is the computer I use most:
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=911510

---

It\'s a moving target. In general, there was a time when Linux was preferred but Windows has the advantage now. Among the offerings for Windows machines it seems there is no clear-cut advantage of one over another though, if I recall correctly, slabs (HadSM3) might be a bit less efficient than the I/O-optimized Coupled Models (HadCM3). The Seasonal Attribution (HadAM3) Models might be the least efficient and they are harder, I think, on the hard disk than Coupled Models -- a consideration for laptops. Personal choice, or flip a coin, eh? Run length influences many people, and the high memory requirement for Seasonal Models could also influence the decision.

In my experience, XP and Vista are about the same, no clear advantage to either one.

Edit: Note that HadCM3 v.6.02 Models have a problem clearing temporary files from the \"dataout\" folder. For now, that\'s a definite consideration when making Model type choice. (We\'re testing a fix on the CPDN Beta site.)
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=6370

---

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.

Edit: Note that HadCM3 v.6.02 Models have a problem clearing temporary files from the \"dataout\" folder. For now, that\'s a definite consideration when making Model type choice. (We\'re testing a fix on the CPDN Beta site.)
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=6370

Actually, you should be able to include HadCM3 in your list of options without any worries. The version offered here on the main project was upgraded to v6.04 about midday on 22 October (the day the project came back to life after the server upgrade), so anyone who started a new HadCM3 since then shouldn\'t hit the problems.

We\'re still keeping a close eye on the new ones just in case, though.....

I heard someone saying he was getting way better results using one of the applications compared to the other ones.

This can refer to 3 different things:
1) The amount of credit this person is getting per unit of time, or
2) His computer keeps crashing one of the model types.
3) Hardware / OS type and it\'s affect on how fast a model runs.

If it\'s 1), then it depends on how many other projects he\'s running, and their types of WUs, as well as the amount of resource share he\'s allocated to each, how much time his computer is turned on but not being used by him, and possibly a few other factors.

If it\'s 2), then that\'s a computer problem.

As for 3), multicore cpus from Intel and AMD handle data differently. AMD (some types anyway), use an on chip memory controller, and have less conflict when several of the models require memory at the same instant.
If this is what yopu\'re talking about, then the way to get the best efficency is to run a different climate model type for each core.
Or to run less climate models than cores, and have a different project\'s WU running on the other core(s).

But ALL of the types of model are being used to produce useful data for different groups of researchers, so all of the different types are needed.

---

Backups: Here

The different model types are all supposed to give similar amounts of credits. So you could calculate how much credit your computer earns on each type of model in, for example, 1 hour or 24 hours. You could then select the type(s) of model the computer performs best with for your next download. Or deselect the type it performs worst with. But each time a new application version is released you\'d really need to run one and check again.

Anyone with a multicore computer should probably run more than one type of model on it. This is also the most interesting and informative way to participate.

From the point of view of producing the maximum possible number of complete models, the best strategy is to look at the CPDN README collections linked in my signature. Choose a backup method, and in the collection about crashes and problems look at item #6 by MikeMars.

---

Cpdn news