Subject: Re: parallel ccm result verfication
From: Erik Kluzek (erik@ncar.ucar.edu)
Date: Fri Sep 17 1999 - 14:02:26 MDT
On Fri, 17 Sep 1999, Akshara Kaginalkar wrote:
> Can anyone help in result verfication between sequential and parallel
> runs of atmospheric models on distributed memory parallel machine
> using MPI.
> 1. does the result from sequential code and parallel version of the
> same code match perfectly ?
Yes, they are bit-for-bit the same. And they are bit-for-bit the same
regardless of the number of processors used in the simulation.
> 3. does the result from sequential runs from two different
> architectures e.g cray and DEC alpha match exactly ?
No, but they match within roundoff of each other. What we do is
to ensure that the "error growth" matches between different machines.
This is done by setting "PERGRO" to true in the LSM namelist. Then
run two simulations setting "pertlim" to 0.0, and 1.e-14 in the CCM namelist.
The RMS difference in temperature between these two simulations should
grow at the same rate for all machines, where the code is correctly
ported.
> 4. Any references for above questions
>
See
http://goldhill.cgd.ucar.edu/cms/ccm3/ccm-users-faq/port.html
For more details on the error growth analysis. Dave Williamson and
Jim Rosinski have published papers on this technique.
Erik Kluzek, (CGD at NCAR)
National Center for Atmospheric Research
Boulder CO, (off) (303)497-1326 (fax) (303)497-1324
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
This archive was generated by hypermail 2b27 : Thu Jun 01 2000 - 09:28:23 MDT