Subject: IBM SP3-compatible version of CCM3.6.6 now available
From: Jim Rosinski (rosinski@cgd.ucar.edu)
Date: Wed Sep 13 2000 - 19:03:18 MDT
CCM Users;
A version of CCM3.6.6 which runs efficiently on clustered architectures such
as the IBM SP3 and Compaq ES40 cluster is now available. At T42 the code can
utilize up to 64 processors, though scaling is reasonably good only up to 32.
On a clustered machine, the number of nodes on which the code is run must
divide evenly into the number of model latitudes.
When run on any of NCAR's SGI Origin machines, the simulation is identical to
that produced with the old CCM3.6.6 tar file provided f90 is used to build
both models. The solution produced has been verified correct to roundoff
error on NCAR's SP3, ES40 cluster, as well as a Linux PC and Sun workstation.
For the moment this code is not configured to run through the CSM
flux-coupler.
Some differences exist in the boundary datasets provided with this
distribution as compared with CCM3.6.6. First, the land model initial
dataset had to be modified because dimension names in the metadata changed.
Thus the LSM initial dataset provided with the CCM3.6.6 distribution will not
work with the new code. Second, a more modern ozone dataset is provided
(noaao3.1990.21999.nc), though the old one from CCM3.6.6 (ozn.0596.r8.nc)
will still work. Boundary datasets suitable for running the code are
included in the source code tar file. So there is no need to retrieve a
separate tar file of datasets.
To get the new tar file, type:
ftp ftp.ucar.edu
login: <anonymous>
cd ccm
bin
get ccm3.10.11.brnchT.366physics.4.tar.gz
Or get it on the web from
ftp://ftp.ucar.edu/ccm/ccm3.10.11.brnchT.366physics.4.tar.gz
For those interested, the rough meaning of the above long name is "CVS tag 4
from an in-house development code (CCM3.10.11) which was modified to be
replaced with CCM3.6.6 physics".
The model now writes its history information directly in netcdf format,
rather than fortran binary as before. It turned out that incorporating this
modification from the in-house development code was critical to achieving
reasonable performance on clustered machines. Since namelist specification
for history tape quantities changed with the new history handling mechanism,
most old namelists based on CCM3.6.6 will not work with the new code. The
land model (LSM) namelist also changed slightly. See the README included in
the tar file for details of the changes.
Users familiar with CCM3.6.6 code will find that that the new code has
changed substantially. The primary reasons for this are 1) we are migrating
to Fortran 90, and 2) the message-passing implementation of the model was
totally rewritten. One consequence of the Fortran 90 migration is that some
files in the distribution end in .F (fixed form), and others end in .F90
(free form). Eventually all source files will be free-form.
Feel free to contact me directly about any problems encountered in building
or running this code. A web-copy of this message is available from the CCM3 News
page at...
http://www.cgd.ucar.edu/cms/ccm3/news/ccm3.6.6_forIBM.shtml
Jim Rosinski
CCM Core Group
rosinski@ucar.edu
This archive was generated by hypermail 2b27 : Thu Jan 04 2001 - 10:02:06 MST