5. FAQ¶
5.1. How do I build and run a single test of the UFS Weather Model?¶
An efficient way to build and run the UFS Weather Model is to use the regression test
(rt.sh
). This script is widely used by model developers on Tier 1 and 2 platforms
and is described in the UFS WM GitHub wiki. The advantages to this approach are:
It does not require a workflow, pre- or post-processing steps.
The batch submission script is generated.
Any required input data is already available for machines used by the regression test.
Once the
rt.sh
test completes, you will have a working copy in your run directory where you can make modifications to the namelist and other files, and then re-run the executable.
The steps are:
Clone the source code and all the submodules as described in Section 3.2, then go into the
tests
directory:cd ufs-weather-model (or the top level where you checked out the code) cd tests
Find a configure (
*.conf
) file that contains the machine and compiler you are using. For this example, the Intel compiler on Cheyenne is used. To create a custom configure file, two lines are needed: aCOMPILE
line and aRUN
line. TheCOMPILE
line should contain the name of the machine and compilercheyenne.intel
and the desiredSUITES
for the build. Choose aRUN
line under thisCOMPILE
command that uses the desiredSUITE
. For example:COMPILE | 32BIT=Y CCPP=Y STATIC=Y SUITES=FV3_GFS_v15p2,FV3_GFS_v16beta,FV3_GFS_v15p2_no_nsst,FV3_GFS_v16beta_no_nsst | standard | cheyenne.intel | fv3 RUN | fv3_ccpp_gfs_v16beta | standard | | fv3 |
Put these two lines into a file called
my_test.conf
. The parameters used in this run can be found in thefv3_ccpp_gfs_v16beta
file in theufs-weather-model/tests/tests
directory.Note
These two lines are long and may not appear in entirety in your browser. Scroll to the right to see the entire line.
Modify the
rt.sh
script to put the output in a run directory where you have write permission:if [[ $MACHINE_ID = cheyenne.* ]]; then stanza: ... dprefix=/glade/scratch
This works for Cheyenne, since
$USER/FV3_RT
will be appended. Also check thatRTPWD
points to a diretory that exists:if [[ $MACHINE_ID = cheyenne.* ]]; then RTPWD=${RTPWD:-$DISKNM/ufs-public-release-20200224/${COMPILER^^}}
Run the
rt.sh
script from thetests
directory:./rt.sh -k -l my_test.conf >& my_test.out &
Check
my_test.out
for build and run status, plus other standard output. Check/glade/scratch/$USER/FV3_RT/rt_PID
for the model run, wherePID
is a process ID. The build will take about 10-15 minutes and the run will be fast, depending on how long it waits in the queue. A message"REGRESSION TEST WAS SUCCESSFUL"
will be written to this file, along with other entertainment:'Elapsed time: 00h:14m:12s. Have a nice day!'
.When the build and run are complete, modify the namelist or
model_configure
files and re-run by submitting thejob_card
file:qsub job_card
5.2. How do I change the length of the model run?¶
In your run directory, there is a file named model_configure
. Change the
variable nhours_fcst
to the desired number of hours.
5.3. How do I select the file format for the model output (netCDF or NEMSIO)?¶
In your run directory, there is a file named model_configure
. Change the
variable output_file
to 'netcdf'
or 'nemsio'
. The variable output_file
is only valid when the write component is activated by setting quilting
to .true.
in the model_configure
file.
5.4. How do I set the output history interval?¶
The interval at which output (history) files are written is controlled in two places, and depends on whether you are using the write component to generate your output files. Table 5.1 describes the relevant variables. If the write_component is used, then the variables listed as model_configure are required. It is however, also required that the settings in input.nml match those same settings in model_configure. If these settings are inconsistent, then unpredictable output files and intervals may occur!
Namelist variable |
Location |
Default Value |
Description |
---|---|---|---|
fdiag |
input.nml |
0 |
Array with dimension |
fhmax |
input.nml |
384 |
The maximal forecast time for output. |
fhmaxhf |
input.nml |
120 |
The maximal forecast hour for high frequency output. |
fhout |
input.nml |
3 |
Output frequency during forecast time from 0 to |
fhouthf |
input.nml |
1 |
The high frequency output frequency during the forecast time from 0 to |
nfhmax_hf |
model_configure |
0 |
forecast length of high history file |
nfhout_hf |
model_configure |
1 |
high history file output frequency |
nfhout |
model_configure |
3 |
history file output frequency |
5.5. How do I set the total number of tasks for my job?¶
The total number of MPI tasks used by the UFS Weather Model is a combination of compute and quilt tasks, and can be calculated using the following relationship:
total tasks = compute tasks + quilt tasks
compute tasks = x layout * y layout * number of tiles
quilt tasks = write_groups * write_tasks_per_group if quilting==.true.
The layout and tiles settings are in input.nml
, and the quilt task settings are in model_configure