[mpas-developers] MPAS Restart I/O

[Petersen, Mark R]

Conrad,

I think separate, individually named restart files would be a big
improvement in the ocean core as well.

Beyond separate names, I would love to see a way to continue from the
latest restart file without specifying the exact time or filename in the
namelist.  That would allow us to line up many jobs, with daisy-chain
dependancies.  

One way to do this is to write the name of the restart file to a text file
named latest_restart whenever a restart file is written.  Then, if the
flags are
config_start_time = 'latest'
config_do_restart = .true.
the code reads the name of the restart file from the latest_restart file.

Now that we are beginning big production runs with mpas-ocean, it is
important to automate the restart process.  We have a similar setup for
POP restarts.

Mark

On 10/25/11 11:50 AM, "Conrad Roesch" <croesch at mail.mmm.ucar.edu> wrote:

>Hello Everyone,
>
>We have made a modification to the restart file I/O for the
>atmospheric cores, and I am wondering how you want to proceed with
>regards to merging these changes back into the trunk.
>
>In the trunk version of the code, multiple restart frames are written
>to a single file, whereas in the atmospheric branch we are writing a
>single frame per restart file, with the filenames differentiated by
>the date/time of the restart frame (for example, if the user specified
>the name "restart.nc" we would write out
>"restart.2011-01-01_12:00:00.nc", "restart.2011-01-02_12:00:00.nc",
>etc).   Our reason for making this change was to avoid some problems
>with overwriting restart frames during subsequent runs and to make the
>available restart options more obvious to the user.  This format will
>naturally make the restart time explicit- that is, it is now
>reasonable to expect the users to specify in the namelist a restart
>time that corresponds precisely to one that is available, and
>therefore in the atmospheric cores we now throw an error for
>non-matching restart times rather than try to interpret them by
>searching for a close available time.
>
>If you are interested in making this the standard way to do restarts
>for the ocean core as well, the changes are straightforward and I
>would be happy to make them.  However, if you would prefer to continue
>writing all restart frames to a single file, I can make the framework
>modifications necessary to continue supporting that method the ocean
>core while employing the one-frame-per-file method for the atmospheric
>cores.
>
>Cheers,
>Conrad
>_______________________________________________
>mpas-developers mailing list
>mpas-developers at mailman.ucar.edu
>http://mailman.ucar.edu/mailman/listinfo/mpas-developers