Introduction
OpenIFS is an ECMWF led project which provides an easy-to-use, exportable version of the IFS system in use at ECMWF for operational weather forecasting. The project aims to develop and promote research, teaching and training on numerical weather prediction (NWP) and NWP-related topics with academic and research institutions.
In this tutorial, OpenIFS has been pre-installed and pre-compiled for you. If you want to learn more about how OpenIFS can be downloaded, configured and installed, please see the OpenIFS website at: http://software.ecmwf.int/oifs/.
In this tutorial...
What you will do:
- Connect to the ECMWF HPC Facility
- Explore installed OpenIFS files
- Run T21 model forecast
- Verify the model works in serial (1 process) and parallel (2 tasks & 2 threads)
- Carry out acceptance test
- Use the model namelist to change settings
- Explore model output data (GRIB)
At the end of this tutorial, you will:
- have learnt about how OpenIFS is installed and organised,
- have run it at T21 resolution and know how to run it serially (1 process) and in parallel with both MPI & OpenMP,
- know how to use grib tools to look at model output.
Connecting to the ECMWF HPC Facilities
In this tutorial, and throughout the entire workshop, you will use one of the Cray XC40 computers which are part of the ECMWF's high-performance computing facility (HPCF).
In this section we will learn how to:
- connect to the HPCF
- start an interactive session on ppn nodes
- Start the Mobaxterm application and open a local terminal
- Login to the RACC using
ssh -X cluster.act.rdg.ac.uk
On the cluster type the command:
ssh troifsX@ecaccess.ecmwf.int
Note: Instead oftroifsX
you should use your ECMWF training user-ID- You will be prompted for the hostname with a choice between
ecgate
,cca
andccb
. You should selectccb
. - When you have completed your work you can disconnect from
ccb
by typingexit
at the command prompt.
The following tasks need to be carried out only once in order to set up the training accounts and the OpenIFS 43r3 model.
- After the first login type the following command:
/home/ectrain/troifs0/setup-cray/my43
- Log out of the training account by typing
exit
Login once more using
ssh troifsX@ecaccess.ecmwf.int
- Type the following command:
get43
The actions above will ensure that your training account receives required scripts and shell configuration files.
The last command will copy the model binaries and input files to your account for version 43r3.
Instead of using the login nodes of the HPCF we will use an interactive session to ask for computing resources and fast temporary disk. This also allows to run parallel jobs in the terminal window.
Type the following command:
% qsub -I -q d
f
-l EC_total_tasks=6 -l EC_job_tmpdir=10G -l EC_memory_per_task=2G
qsub: waiting
for
job 7215630.ccbpar to start
qsub: job
7215630
.ccbpar
ready
The changed command line prompt indicates that we have switched from the login node to a pre/post-processing node.
Important: After completing your work you need to close the interactive session by typing exit
which will bring you back to the login node.
On the ECMWF HPCF an interactive session will last for 48 hrs by default, unless the walltime is specified using an additional directive.
OpenIFS directories
In this section we:
- Set the OpenIFS environment
- Examine the OpenIFS installation
- Carry out the tasks above to connect to
ccb
and start an interactive session - Change into the model's main directory:
cd perm/oifs43r3
- Type the command:
source ./oifs-config.ccb.sh
The oifs-config.ccb.sh
script sets a number of Unix shell environment variables which define the type of OpenIFS compiled installation and location of files. These settings are specific to version 43r3.
If you do not run this setup script (or alternatively include its content in your Unix shell configuration file) then OpenIFS will not run correctly.
For more information on how to configure OpenIFS, please see the OpenIFS website: http://software.ecmwf.int/oifs/.
- Make sure you are in the directory:
~/perm/oifs43r3/
- List the files in this directory using the
ls
command
The directory contains a number of text files and directories:
% ls bin/ COPYING LICENSE oifs-config.ccb.sh READMEs/ CHANGES examples/ make/ python/ src/ CITE fcm/ NOTICE README t21test/
Directories:
src
- contains all the source code for the model and supporting programs.make
- contains the build configuration files for the FCM compile command. Object files and executables will be in this directory organised according to the type of build (OIFS_BUILD environment variable).t21test
- self contained T21 model run for verifying the installation is correct.ifsdata
- contains additional input files for the model e.g. climatologies. Available as separate tarfiles.
Note:
- OpenIFS has been precompiled on the HPCF.
- All source code has been removed due to licensing restrictions.
- OpenIFS builds 'out-of-source'; this means object (
.o
) files and executables (binary files) are not mixed with the source code. - The
README
file contains information about software requirements, setting up the local compilation environment, and where to get help and support.
OpenIFS T21 test forecasts
In this section of the tutorial, we will run the pre-compiled OpenIFS model on a simple T21 forecast.
You will:
- learn about OpenIFS input and output files.
- learn some switches to control OpenIFS.
- learn how to run the model in parallel.
- learn how to run the acceptability test.
- Make sure you are in the directory: ~/perm
/oifs43r3/t21test
- List the files in this directory.
The directory t21test
contains a number of files:
% ls t21test ICMGGepc8INIT ICMSHepc8INIT job ref_021_0072 ICMGGepc8INIUA ifsdata/ namelists run.ppn
Files beginning with 'ICM'.
These are the input files for this T21 experiment. They are in GRIB format. Do not move them from this directory. OpenIFS expects to find its input files in the same directory as the main executable.
epc8
- this is the Experiment ID. Experiments IDs are used at ECMWF and initial conditions provided by ECMWF will always have an expt id.ICMGGepc8
- 'GG' indicates these contain gridpoint fields.ICMSHepc8
- 'SH' indicates these contains spherical harmonic fields.
job
Shell script to run the model. Described in more detail below.
run.ppn
Simple shell script which calls job in an interactive shell environment.
ifsdata
Climate data fields used for T21 test integration. You should not move or rename this directory as the model will expect to find the climate files it needs in a directory of this name.
namelists
This file contains all of the input model fortran NAMELISTS. Not all of the namelists have their variables listed, only the variables commonly changed are listed here. Users should copy this file and modify it for the tests described below.
ref_021_0144
This file is reference output for the model tests. The model can be run in 'reference' mode where it checks it is working correctly by comparing some mathematical norms against these files. Reference runs are described in more detail under 'Acceptance testing' below.
Use the grib_ls
and grib_dump
commands to examine the contents of the ICM files.
Run the model:
% ./run.ppn
What happens?
The model fails. Look at the standard output (or in the NODE_001.01
file when it is created) and find the subroutine traceback. Near the top of the traceback you will find the error messages.
Whenever the model fails, it will produce this traceback (controlled by DR_HOOK=1 in the job
file).
Single process test
Copy the file namelists
to fort.4
and run the model with a single task and single thread by executing the job script:
% cp namelists fort.4 % ./run.ppn
The model will expect to find a file called fort.4
in the same directory as the executable. This script copies the executable from make/cce-opt/oifs/bin
.
If the run works you will see output like:
... 17:09:08 STEP 1 H= 0:10 +CPU= 0.276 STEP 1 :## EC_MEMINFO 1 ccbppn01 309 206 0 16530 190 16080 1188 35807 61171 0.2 0.0 s/p 17:09:08 STEP 2 H= 0:20 +CPU= 0.280 STEP 2 :## EC_MEMINFO 1 ccbppn01 309 206 0 16530 190 16011 1188 35740 61170 0.2 0.0 s/p 17:09:08 STEP 3 H= 0:30 +CPU= 0.268 STEP 3 :## EC_MEMINFO 1 ccbppn01 309 206 0 16530 190 16008 1188 35734 61170 0.2 0.0 s/p 17:09:09 STEP 4 H= 0:40 +CPU= 0.264 STEP 4 :## EC_MEMINFO 1 ccbppn01 309 206 0 16530 190 15966 1188 35695 61170 0.2 0.0 s/p 17:09:09 STEP 5 H= 0:50 +CPU= 0.268 STEP 5 :## EC_MEMINFO 1 ccbppn01 309 206 0 16530 190 16008 1188 35734 61171 0.2 0.0 s/p 17:09:09 STEP 6 H= 1:00 +CPU= 0.004 STEP 6 :## EC_MEMINFO 1 ccbppn01 309 206 0 16530 190 16008 1188 35734 61171 0.2 0.0 s/p
This test runs only a small number of timesteps.
Model output
The model writes its output to a several files.
NODE_001.01
contains the text output (WRITE/PRINT statements). The numbers refer to task number and thread number. Only output from the master task & thread is normally output but this can be changed for debugging purposes.
ICM*epc8+0000
is the model output in GRIB format split into 2 files; one for the gridpoint, the other for spectral fields. These contain only a few output variables in this test. This file is a mix of GRIB1 and GRIB2 messages. See the Documentation for how to process this output.
ifs.stat
is a small file that prints the model steps, time taken for each step and a 'norm' measure. This file can be usually ignored but is useful for debugging.
Look at the output from the model in the NODE_001.01
file from this successful run. Note the output of the model namelists and the statistics printed at the end. IFS has very comprehensive logging output which is useful for debugging and understanding the model's performance.
Look at the grib output files using the grib commands.
Parallel runs
These next short tests verify the model works correctly with either OpenMP parallel threading, MPI tasks and both and follow on from the serial tests above.
Edit the file run.ppn
and change the line: export NTHREADS=
1 to NTHREADS=2
Run ./run.ppn
as above.
Do the reported CPU times change?
Use the grib_ls
command to look at grib output files - what do you notice?
OpenMP threads is only enabled for optimized 'opt' builds
If this works, look in the NODE_001.01
output file for the line:
NUMBER OF THREADS 2
to verify the model ran with 4 OpenMP threads.
Edit the file run.ppn
and change NTHREADS
back to 1.
Change the line: NPROC=1
to NPROC=2
.
Also, edit the fort.4 file and change NPROC
to 2.
Rerun the job:
./run.ppn
Do the reported CPU times change?
Note that increasing the number of tasks requires changing the number of tasks in two places.
Look in the NODE_001.01
output file for the line: "NPROC = 2
" to verify that two MPI tasks was used.
Mixed mode: OpenMP/MPI
The model can also be set to use NPROC=2
and NTHREADS=2
to use a total of 4 processes. However, this would require a computer with at least 4 cores.
Acceptance testing
The final step is to check the model is producing the numerical answers within acceptable limits, even if it runs the short tests above without failing. OpenIFS includes code that will compute internal statistical norms and compare against numbers supplied by ECMWF. The file: ref_021_0144
in the t21test
directory contains statistical norms computed by the model run at ECMWF.
To do the acceptance test, edit the namelists in fort.4
and look for the NAMCT0 namelist:
&NAMCT0 LREFOUT=false
set the variable LREFOUT to TRUE:
&NAMCT0 LREFOUT=true
With LREFOUT=true
, at the last timestep OpenIFS will read the ref_021_0144 file and produce a new file: res_021_0144 (note the similar filenames!). The contents of the file should be similar to:
% cat res_021_0144 Results of ERROR calculation The error calculated from the results shows that the calculations are correct The maximum error is = 0.11345 %
As long as the model reports 'calculations are correct' and the error is less than a few percent then the model is behaving satisfactorily in your compilation and run environment.
How to control model output
In this section, the main NAMELIST variables that control the output of the model are presented.
If you have time, try changing the variables, run the model and using the grib_ls and grib_dump commands to view the output grib file contents.
How to control output frequency
The namelist variables that determine the output from the model as it runs are:
Namelist : NAMCT0
NFPOS
- this should be set =2 in order to turn on model output and diagnostics.
NFRHIS
- this is the output frequency of the 'history' files, that is, the model's prognostic variables on the model levels.
- this is the output frequency of the prognostic variables on pressure levels.
NFRPOS
It's recommended these are set the same.
If NFRHIS/NFRPOS
are positive, the units are in model timesteps. If a negative value is used, the units are in hours.
NPOSTS & NHISTS - these are integer arrays that control the write times of the history files. They can be used for non-regular output intervals.
Examples
Regular output at fixed timesteps
NFRHIS=4,
NFRPOS=4,
NPOSTS=0,
NHISTS=0,
This simple example will cause the model to produce history file output every 4 timesteps.
For this to work correctly, NFRHIS * timestep
must equal an integer number of hours. The GRIB output will not work correctly if this isn't the case.
Non-regular output
NFRHIS=1,
NFRPOS=1,
NHISTS(0)=3,
NHISTS(1:3)=0,-3,-9,
NPOSTS(0)=-3,
NPOSTS(1:3)=0,-3,-9,
The minus sign indicates the units are in hours rather than timesteps. NFRHIS/NFRPOS
in this case must be set to 1. The 0th element of NHISTS/NPOSTS
determines how many outputs are produced in total by the model, the first to nth elements determine the actual output times (hours in this case because of the negative values used).
In this example, the model will write 3 separate output files at the first timestep (0hrs), 3hrs and 9hrs and then no more regardless of how long the model runs for.
How to change the output variables and post-processing
The namelist NAMFPC
is the main namelist for the post-processing. Variables in this list can be sensitive to changes as many combinations are possible but not all work.
Model level output
To control model level output the following namelist variables (in NAMFPC) are used:
NRFP3S - list of the model levels on which post-processed output is required.
e.g. for a 60 level model run where output on all levels was required set:
NRFP3S=1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,
For a 91 level model, this would give output on the first 60 levels (starting from the top).
NFP3DFS - number of 3D fields to be output on model levels. Must equal number of entries in MFP3DFS.
MFP3DFS - list of grib codes of 3D variables to be output on model levels. See How to control OpenIFS output on the OpenIFS website for valid codes.
e.g
NFP3DFS = 5,
MFP3DFS = 130, 135, 138, 155,
would output the temperature (130), vertical velocities (135), relative vorticity (138), divergence (155) on model levels.
Using the information above, try:
- Adjusting the output frequency of the model.
- Changing the list of model levels used for output and the output variables.
At the end of the session
Do not forget to close your interactive ppn session on the HPCF using the exit command.