GEMCLIM tools (for UQAM users)
A little reminder:
If you reach a collaboration web page asking
for login and password information, try using 'science' for both.
Batch jobs
On 'marvin', 'st1', 'colosse' and Linux machines/clusters in Dorval the
Sun Grid Engine (SGE) queueing system is used.
Submit a job
If you want to submit a job on marvin, st1 or on any Linux
machien/cluster in Dorval use 'soumet'.
On marvin
please always
use the 'soumet' from the environment and never
use 'qsub' directly since this
might disturb our
queuing
system!
Soumet is a script that prepares a job to be submitted by inserting
appropriate Korn shell or Posix shell commands before and after the
user commands that make up the job.
The command to submit a job (script) could look like this:
soumet
jobname -t time_in_seconds
-listing listing-directory -jn listing_name [-cpus number_of_cpus]
Check on jobs
To check which jobs are running/queued use 'qstat'.
On marvin, st1 and colosse you can also use my
little script '~winger/ovbin/qs'.
Kill a job
Check the job-ID number with 'qstat' or '~winger/ovbin/qs' and then use
'qdel' to
kill your job:
qdel
job-ID
Batch queues
marvin
On marvin there are 48 compute nodes and 1 headnode. Each of them
having 2 dualcore processors. Which makes 192 cores
to
run the model and other batch jobs. The maximum number of cores used
for OpenMP is 4.
At the moment there are two batch queues on 'marvin'.
One queue for jobs requesting 1
or 2
cpu's, another queue for jobs requesting a multiple of 4 cpu's.
Jobs asking for a number of cpu's which is higher than 4 and not a
multiple of 4 are not allowed.
There are no classifications or restrictions related to the runtime
requested.
st1/st2/st3
st1/st2/st3 are 3 nodes, each of them with 2 quadcore processors. Which
makes 24 cores
in total. The maximum number of cores used for OpenMP is 8.
Interactive jobs on marvin
Interactive jobs should only be submitted from node cn149,
never
from the headnode!
You can logon to the node with:
ssh
cn149
Creating
an executable
r.compile
compiles sources in Fortran (.f, .ftn, .ptn, .ftn90, .f90, .cdk90)
and/or in C (.c) with all the needed options to insure compatibility
with the corresponding librmn library and build (on demand) an
executable file for the specified platform. Unless really
specific
needs, few compilations options
(key -opt) will have to be specified since the use of the
proper
keys
will adjust the compilers accordingly (paths for libraries and included
files,
particular flags for the compilers, etc...).
By default, the script will compile for the platform on which the
script is called.
Usage example:
r.compile -src
source -o executable
-librmn
The key -librmn include the library 'librmn' which includes i.e. the
fst-functions.
r.build
links objects (.o) and create an executable with all
the needed options to insure compatibility with the corresponding
librmn
library for the specified platform. Unless really specific
needs,
few
compilations options (key -opt[f,c]) will have to be specified since
the use of
the proper keys will adjust the compilers accordingly (paths for
libraries
and included files, particular flags for the compilers, etc...).
By default, the script will compile for the platform on which the
script is called.
cmc-archive
A cmc-archive files is very similar to a UNIX tar-file. The extension
of a cmc-archive file is always '*.ca'.
To get more information simply type:
cmcarc -help
Look at a cmcarc:
cmcarc -t -v
-f archive
Extract files from a cmcarc:
cmcarc -x [file_list] -f
archive
Wildcard are allowed for 'file_list'. I.e. '*.'.
Add to a cmcarc:
cmcarc -a file_list -f
archive
Wildcard are allowed for 'file_list'. I.e. '*'.
Author: Katja Winger
Last update: January 2010