[Tgcmgroup] new tgcmproc
Ben Foster
Ben Foster <foster@hao.ucar.edu>
Thu, 30 Jan 2003 11:39:21 -0700 (MST)
Dear tgcmgroup:
I am releasing an updated version of tgcmproc (tgcmproc1), our fortran
post-processor for the tiegcm and time-gcm models. This code has been
ported to the IBM platform. For the next couple of weeks tgcmproc1 will
be available for testing on the IBM's and the SGI (the old version will
still be available on the SGI dataproc). After this test period, I will
make tgcmproc1 the default on all machines.
Executables are stored on the ncar mss as follows:
/TGCM/tgcmproc/tgcmproc_aix # executable for IBM
/TGCM/tgcmproc/tgcmproc1_irix # executable for SGI
/TGCM/tgcmproc/tgcmproc_o2k # old tgcmproc for SGI
The new source code is at hao /home/tgcm/tgcmproc/tgcmproc1, and also on
the mss as /TGCM/tgcmproc/tgcmproc1.tar.Z
Job scripts to make batch runs of tgcmproc1 are at hao:
/home/tgcm/tgcmproc/tgcmproc_ibm.job
/home/tgcm/tgcmproc/tgcmproc_sgi.job
!-----------------------------------------------------------------------
To get started with tgcmproc1 on dave (a special post-processing and
visualization node on blackforest) or bluesky, take the following steps:
1. Copy the job script /home/tgcm/tgcmproc/tgcmproc_ibm.job to your own
directory, make it writeable by owner (chmod 644), and make the following
changes:
a. Change REMOTE to your own machine:path to receive output listings.
b. Change "tgcmproc.inp" to your own namelist input file (input files
used with the old tgcmproc should work with tgcmproc1). Input files
given in the INCLUDE statement can be a local file, or full or relative
paths (e.g., ~/tgcmproc/tgcmproc.inp).
c. Set the class (IBM's name for queue) to "dv_reg" (or "share") for
dave, or "share" for bluesky.
# @ class = dv_reg
or
# @ class = share
d. Set the wall_clock limit. Setting this as low as possible will reduce
time spent waiting in the queues. You will learn from experience
how low you can set it for different runs. The wall_clock limit
on both machines is 6 hours.
2. Make the appropriate entries in your .rhosts files on bluesky, dave,
and your hao workstation (note that dave and blackforest share the
same home). Make sure that rcp works in both directions between
all 3 machines (contact me at x1595 or the consultants at x1278 if
you need help with this).
3. Put the following lines in your .cshrc files on dave and bluesky:
setenv NCARG_LIB /usr/local/lib32/r4i4
setenv NCARG_ROOT /contrib # on dave (/usr/local is also ok on dave)
setenv NCARG_ROOT /usr/local # on bluesky
4. Submit to dave or bluesky using the "submit" command (/home/tgcm/bin/submit):
submit tgcmproc_ibm.job.
5. Note that if you already have model history files on the IBM disk
(e.g., from a recent model run) you can set the namelist input "tmpdir"
to this directory so tgcmproc will not have to go to the mss. (You can
run tgcmproc concurrently with the model using this method).
!-----------------------------------------------------------------------
To run interactively on dave or bluesky, take the following steps:
(after fixing .rhosts and .cshrc files as in steps 2 and 3 above)
1. Login to either dave or bluesky, go to a directory under /ptmp, and
get the tgcmproc1 executable from the mss with the following command:
msrcp mss:/TGCM/tgcmproc/tgcmproc_aix .
2. Give executable permissions:
chmod u+x tgcmproc_aix
3. Copy your namelist input file from your hao workstation using rcp.
4. Execute with your input file:
tgcmproc_aix < tgcmproc.inp
!-----------------------------------------------------------------------
Running tgcmproc1 with 2.5 degree resolution histories requires recompilation
with modification to a single source file. To do this:
1. Copy the job script /home/tgcm/tgcmproc/tgcmproc_ibm_2.5h.job
2. Change REMOTE and tgcmproc_2.5h.inp
3. Submit with "submit tgcmproc_ibm_2.5h.job"
Note if you exceed your home disk space with this command, copy the
job script to a /e or /d directory and resubmit (you can INCLUDE the
full path to your input file).
!-----------------------------------------------------------------------
To run tgcmproc1 on SGI dataproc, use the following job scripts:
/home/tgcm/tgcmproc/tgcmproc_sgi.job
/home/tgcm/tgcmproc/tgcmproc_sgi_2.5h.job
!-----------------------------------------------------------------------
Highlights of the new version:
o Tgcmproc1 runs on both SGI and IBM platforms (can be run on NCAR's SGI
dataproc, IBM dave (a special purpose node on blackforest), and on
IBM bluesky (interactive or share queues).
o Interpolation of tgcm model results to satellite observational tracks
e.g., TIMED mission SABER and TIDI orbit tracks (Craig Hartsough code
additions to tgcmproc).
o New netcdf output file, including 2 new namelist options to customize
the vertical dimension (zp or ht) of fields written to the output file.
o Processing of 2.5 degree resolution model histories by recompilation
with modifications to a single parameter file (proc.f).
o Processing of Jupiter (jtgcm) histories.
o Several bug fixes, new derived fields, repaired xdr output.
o Anti-highlight: tgcmproc1 still does NOT process fields on the geomagnetic
grid (maybe next time).
Please see /home/tgcm/tgcmproc/tgcmproc1/README for all the gory details.
(see especially the notes on netcdf output files)
!-----------------------------------------------------------------------
Please let me know if you have any problems with tgcmproc1 during the
test period. I discourage everybody from becoming dependent on modifications
to the source code (with the exception of 2.5h runs). If you are tempted to
modify the source, let me know and we can discuss the problem. This will
minimize the impact to users of updating the code in the future. Good Luck!
--Ben
-----------------------------------------------------------------------
Ben Foster High Altitude Observatory (HAO)
foster@ucar.edu phone: 303-497-1595 fax: 303-497-1589
Nat. Center for Atmos. Res. P.O. Box 3000 Boulder CO 80307 USA
-----------------------------------------------------------------------