Re: Model Timestep


Subject: Re: Model Timestep
From: Peter Paul Smolka (smolka@uni-muenster.de)
Date: Thu Aug 03 2000 - 14:52:55 MDT


On Thu, 3 Aug 2000, John B. Eylander wrote:

> Date: Thu, 3 Aug 2000 14:31:21 -0400
> From: John B. Eylander <bj@thor.gsfc.nasa.gov>
> To: "Ccm-Users (E-mail)" <ccm-users@UCAR.EDU>
> Subject: Model Timestep
>
> Hey all,
>
> I have a question about the model timestep of CCM3.6. After working my way

( ... )

Hi John,

the warning about changing the time-step you have seen.

Thus you raised the question about the computer:

If I understood you correctly:

> model that I could run locally and generate long period datasets (100 years
> or so). Using CCM, I performed a test run of 100 days which took about 24
> hours on my SGI OCTANE, with dual R12000 300 mHz processors. This seems

         100 days => 24 hours
          10 days => 2.4 hours 120+0.4*60 = 120+24= 144 min
           1 day => 14.4 min.

You reach the speed of an semi-fast office PC:

Pentium II/300, 66MHz Bus: 1 Day: 28 Min (Dell, 3 years old)
IBM Aptiva /300MHz, unknown bus: 1 day: 18 min (as reported)

So maybe some Pentium 933 with fast bus, noble components and selected
compiler options should help (see below: Cray speed).

> model (at least to me). At this rate, I figure a 100 year climate run would
> take nearly 1 year to calculate!

A well planned experiment with restart runs every year (and burning
the interim results on CD) frees you while the PC is running
for other work.
>
> Is there something I am missing here, or am I correct is assuming that long
> term runs (100's of years) should not be accomplished on my desktop SGI, and

s.a.: A fast PC should beat SGI easily.

The NCAR manual for ccm3.2 (not 3.6) said in ONE processor operation
(which is unusual) the Cray needed 9 minutes per day. I found the ccm3.6
is as fast as the 3.2. although it delivers more (LSM etc.).

And: With a fast PC you might (bus speed and other factors)
reach a Cray in one processor mode (256 MB RAM of course).

365 x 9 x 100 = (too lazy to take a pocket calculator) =

> only run on a much more powerful Cray? Or am I able to change the timestep
> to something a little more do-able, on the order of a few hours (I see in

A few hours I regard unrealistically and parallel computing
is not as easy as "just setting some compiler options".

(so compare the time "adjusting the model incl. error tracking"
with "running one or more PCs in your room").

Maybe, if you have dual processor systems you check the KAP
preprocessor which exists for various environments.
(I suspect, considering the structure of ccm3.6 with above
a maximum gain of factor 1.5 to 1.6 is achievable. Considering
the risks (and the costs of your worktime): A faster PC is cheaper.

> the documentation the warning about changing the timestep, but I figured I'd
> pose the question anyway)? Should I be looking for a different climate
> model instead? Thanks for any answers you can throw this way!
>
Maybe, as said, you look for a different PC as
(I am NOT a fan of a certain company) often
Unix environments and workstations have been overstimated.

Different climate model: It depends on your question what you need
(= not on the computer speed).

If a Pentium is NOT misused for "click and wait" but operated
in command mode,it really shows the tremendous performance gains
that have been achieved in the past years.

Hope that helps (as you mentioned SGI as bottleneck).

Best regards, Peter

> John Eylander
>
>
>

**********************************************************************
Dr. Peter P. Smolka
University Muenster
Geological Institute
Corrensstr. 24
D-48149 Muenster

Tel.: +49/251/833-3989 +49/2533/4401
Fax: +49/251/833-3968 +49/2533/4401
E-Mail: smolka@uni-muenster.de
E-Mail: PSmolka@T-Online.de
**********************************************************************



This archive was generated by hypermail 2b27 : Thu Jan 04 2001 - 10:01:58 MST