Dalhousie Geodynamics Group

Run

On the scrXm family, the stable, and geodyncomp

• ssh to a sopale-enabled computer.
• Change to the directory where you want to do this run.
• Prepare an input file, sometimes called sopale_nested_i, using your favourite text editor. Later versions of sopale_nested will accept the name of the input file as a parameter. For those versions, any filename will do.
• Select a version of sopale_nested from:
  • ls -ld /home/dguptill/software/sopale_nested*
• Select a build from:
  • ls -l /home/dguptill/software/sopale_nested-<version>/bin
• Start it, using the 'start' script for the build size you have selected:
  /home/dguptill/software/sopale_nested-<version>/bin/start_sopale_nested-<build>
• The script usually requires parameters. When invoked without parameters - as in the example above - it will print a message describing the parameters required.

Run-Multi-Phase

This feature has changed, and may change again, as sopale_nested evolves. See the /bin directory of any sopale_nested version. Run the script without parameters to get the user documentation.

• ls -l /home/dguptill/software/sopale_nested-<version>/bin

Restart

sopale_nested generates restart files at periodic intervals, and has the ability to continue a model from them.

In order to restart sopale_nested, you will need

1. an input file for sopale_nested. The input file from the original run can be used, with the following changes:
   • "rres" needs to be set to 1.
   • If "align_time" = 1 then the model will restart at the timestep and time in the restart files.
   • If "align_time" = 0 then the model will restart with timestep = 1, and time = 0.
   • Change "nameout1" to match "nameout1" in the restart file names.
   
   
2. restart files. To make the restart files:
   • rename any matching set of output files
     1. <nameout1>g01_p00_T<ff>_o
     2. <nameout1>g01_p00_f<ff>_o
     3. <nameout1>g01_p0n_T<ff>_o (one of these for each nest, where 'n' varies from 1 up)
     4. <nameout1>g01_p0n_f<ff>_o (one of these for each nest, where 'n' varies from 1 up)
     5. <nameout1>g02_p00_T<ff>_o
     6. <nameout1>g02_p00_f<ff>_o
     by changing
     • <nameout1> to the new run_name
     • <ff> to 00
     • '_o' to '_i'
   • There is a script which automates these name changes. Look for 'make_restart_files' in
     /home/dguptill/software/sopale_nested-<version>/bin

Version 2011-09-06_2.0 and earlier do not require the header files. (The files whose names end in _T00_o)

Kill

On occasion, it may be desirable to stop a model before it has completed.

Is It Still Running?

Sometimes, such as when sopale_nested can no longer write output files due to quota issues, or because one of the four processes involved in a one-nest job has been killed, the number-crunching processes remain running at 100%, using cpu cycles, but producing no visible results. Based on recent experience, this happens more often than is desirable.

Locating such errant jobs and/or processes can be done by running a script, and filtering the output for one's username. The output includes the date on which the processes were started. That date can be an indicator of a 'lost' job, if it is more in the past than one would expect for a sopale_nested run.

Since version 2009-07-26_29.0

The command to find the pid of your job is in the file

• <nameout1>.out

• kill <pid>

It is often desirable to use disc space on one server and the cpu of another. To use the cpu of scrom, and disc space on scrim:

1. login to scrom
2. cd
3. fr &
4. mkdir -p scrXm/scrim
5. sshfs scrim.ocean.dal.ca:/ scrXm/scrim
6. cd scrXm/scrim/<some directory>
7. run sopale_nested

You only have to do Steps 4 and 5 once. If either scrom or scrim is rebooted, the directory scrXm/scrim will be empty. Repeat step 5 to fill it.

Mix-And-Match (geodyncomp)

dg322872 wants to use the cpu of sopale-0-1, but discovers he has no quota there. However he has quota on sopale-0-3 in /state/partition2. What to do?

# ssh to sopale-0-1
ssh -X sopale-0-1
# [optional: start filerunner]
/share/apps/fr &
# make a place to mount the disc of sopale-0-3
mkdir -p geodyncomp-nodes/sopale-0-3
# mount the disc of sopale-0-3
sshfs sopale-0-3:/ geodyncomp-nodes/sopale-0-3
# cd to the directory he wants to use
cd geodyncomp-nodes/sopale-0-3
cd state/
cd partition2
# The first time, he will have to create this directory
# mkdir dg322872
# The directory already exists, so he just cd's
cd dg322872
mkdir -p testing/sopale_nested
cd testing/sopale_nested
# get a copy of the sopale_nested input file
cp -ip /home/dg322872/testing/sopale_nested/sopale_nested_i.Guilmette-makeAnOcean-P1_DIV-20120608_2.0 .
# run sopale_nested (line broken for better formatting)
/home/dguptill/software/sopale_nested-2012-06-08_2.1/bin/\
  start_sopale_nested-2012-06-08_2.1-ompimod-intel-lib-11.0.074-std_large \
  2 no-email nosync sopale_nested_i.Guilmette-makeAnOcean-P1_DIV-20120608_2.0 

Mix-And-Match (scrXm and stable to geodyncomp)

It is possible to use the cpu of scrym, and the disc space on a geodyncomp node. The geodyncomp nodes cannot be seen from the outside world, so a two hop strategy is required. Make the sshfs mounts from the scrXm (or stable) machine to geodyncomp to the geodyncomp node.

# login to scrym
cd
mkdir -p scrXm/geodyncomp
sshfs geodyncomp.ocean.dal.ca:/ scrXm/geodyncomp
ssh -X geodyncomp.ocean.dal.ca
mkdir -p geodyncomp-nodes/sopale-0-0
sshfs sopale-0-0:/ geodyncomp-nodes/sopale-0-0
# create another terminal on scrym
cd scrXm/geodyncomp/geodyncomp-nodes/sopale-0-0/<some directory>
run sopale_nested

This page was last modified on Thursday, 30-May-2013 14:15:21 ADT
Comments to geodynam at dal dot ca