wiki:Doc/Models/DYNAMICO

DYNAMICO


1. Introduction

The DYNAMICO project develops a new dynamical core for LMD-Z, the atmospheric general circulation model (GCM) part of IPSL-CM Earth System Model. The primary goal of DYNAMICO is to re-formulate in LMD-Z the horizontal advection and dynamics on a icosahedral grid, while preserving or improving their qualities with respect to accuracy, conservation laws and wave dispersion. In turn, a new grid refinement strategy is required. A broader goal is to revisit all fundamental features of the dynamical core, especially the shallow-atmosphere/traditional approximation, the vertical coordinate and the coupling with physics. Efficient implementation on present and future supercomputing architectures is also a key issue addressed by DYNAMICO.

2. Getting DYNAMICO

DYNAMICO is licensed under the CeCILL open source license. The latest version is accessible through svn :

svn co http://forge.ipsl.jussieu.fr/dynamico/svn/codes/icosagcm/trunk

Registered IPSL forge users belonging to the DYNAMICO group can do :

svn co svn+ssh://mylogin@forge.ipsl.jussieu.fr/ipsl/forge/projets/dynamico/svn/codes/icosagcm/trunk DYNAMICO

'mylogin' should be your forge login. svn will create the DYNAMICO directory and download the source code there. Source can also been browsed at https://forge.ipsl.jussieu.fr/dynamico/browser/codes/icosagcm/trunk

3. Compiling DYNAMICO

DYNAMICO is written in Fortran 90 with some legacy code in Fortran 77. The build process is based on FCM. DYNAMICO requires the NetCDF library, including the F90 modules. MPI is required for parallel execution but DYNAMICO can compile and run without MPI.

DYNAMICO depends on the NetCDF and BLAS libraries.

The compiling process is automated but some information is needed to guide it. This information is contained in text files in https://forge.ipsl.jussieu.fr/dynamico/browser/codes/icosagcm/trunk/arch . Sample files corresponding to a few machines (Jean-Zay at IDRIS, Irène at TGCC) are present. Assuming you compile on Irène :

cd DYNAMICO
./make_icosa -arch X64_IRENE -parallel mpi -prod -job 8

will compile DYNAMICO. The make_icosa script accepts keyword-value pairs which drive its behaviour. The most important option is arch=ARCH (here ARCH=X64_IRENE). It directs make_icosa to use the information contained in :

  • arch/arch-ARCH.env
  • arch/arch-ARCH.fcm
  • arch/arch-ARCH.path

The *.env file is a shell script that is executed by make_icosa. It sets up the environment for use by *.path. *.path defines paths to libraries and modules needed for compilation. *.fcm defines the commands used to compile, link, etc. as well as options to be passed to the compiler/linker .

The option "-job 8" is similar to "make -j 8" and compiles in parallel for speed. After a successful build the main executable is found in the bin/ directory.

Compiling with XIOS output

DYNAMICO can direct its output through XIOS, a parallel I/O library and server. See https://forge.ipsl.jussieu.fr/ioserver. To enable XIOS output :

  • get and compile XIOS in a separate directory
  • set the variables XIOS_INCDIR , XIOS_LIBDIR and XIOS_LIB to appropriate values in your arch.path
  • use the option "-with_xios" in your "make_icosa" command

Why use XIOS :

  • without XIOS, each output field is written to a separate NetCDF file. Post-processing is required to group several fields together.
  • with XIOS, several fields can be written to a few output files. This behavior is controlled by the input file xios.xml (required). See https://forge.ipsl.jussieu.fr/ioserver for the syntax of this XML file.
  • without XIOS, data to be written is communicated to the main MPI process, which writes to the NetCDF files while other MPI processes stay idle. This is not expected to scale to a large number of MPI processes.
  • XIOS provides asynchronous, parallel I/O in order to scale to large MPI process counts.

4. Running dynamico

DYNAMICO can produce a rather large amount of output. It is therefore recommended to prepare a separate directory for each numerical experiment on a filesystem of adequate capacity. In this directory, copy the executable icosa_gcm.exe. You will find it in DYNAMICO/bin/, where DYNAMICO is the main directory containing the source code.

You will also need configuration files that define resolution, initial condition, etc. Sample files can be found in subdirectories of https://forge.ipsl.jussieu.fr/dynamico/browser/codes/icosagcm/trunk/param_sets. Copy the *.def files from the desired sub-directory. There is typically run.def and earth_const.def . run.def is the main configuration file and includes earth_const.def

Currently DYNAMICO generates its own grid. It can run idealized test cases which define their own initial conditions. In that case no other input files are needed beyond *.def and, if using XIOS, *.xml files controlling XIOS behavior. It may also restart from a previous run, reading from a restart file.

Running dynamico on parallel computers

DYNAMICO can run in parallel by dividing the icosahedral mesh into tiles. There are at least 10 tiles corresponding to the 20 faces of the icosahedron joined in pairs to form rhombi. These 10 rhombi are further subdivided in nsplit_i x nsplit_j tiles. nsplit_i and nsplit_j are defined in run.def . nsplit_X needs not divide nbp exactly.

To run DYNAMICO on a parallel machine, you must first compile it with OpenMP and/or MPI. Then use mpirun or the equivalent command to run it and OMP_NUM_THREADS environment variable to fix the number of openMP threads. There must be less MPIxOpenMP processes than the 10 x nsplit_i x nsplit_j tiles. There can be more tiles than processes, in which case each process will take care of several tiles.

Last modified 2 years ago Last modified on 10/11/19 16:04:37