Suggestions for software installation on Betzy after upgrade

[TomasTorsvik]

Hi,

looking at the available modules on the test node login-3.betzy.sigam2.no, it seems that ncview is not supported at the moment. There may be other things missing as well. I would like to send a request to NRIS to install software that we need, to have this available as a consistent set of modules (i.e. without needing to swap out GCCcore to load a specific tool-chain).

I'm stating a list of software that I think can be useful to have, please add in this discussion

CDO
git
Mamba
Miniconda (not sure if this is needed in addition to Mamba)
ncview
netCDF
netCDF-Fortran
PnetCDF
Python

---

Update: 2024-09-17
At the moment we have most tool-chains available in versions consistent with 2022a and GCCcore/11.3.0

GCCcore/11.3.0
CDO/2.0.6-gompi-2022a
ESMF/8.3.0-foss-2022a
ESMF/8.3.0-intel-2022a
ESMF/8.3.0-iomkl-2022a
git/2.36.0-GCCcore-11.3.0-nodocs
HDF5/1.12.2-gompi-2022a
HDF5/1.12.2-iimpi-2022a
HDF5/1.12.2-iompi-2022a
NCO/5.1.9-iomkl-2022a
ncview/2.1.8-gompi-2022a
netCDF/4.9.0-gompi-2022a
netCDF/4.9.0-iimpi-2022a
netCDF/4.9.0-iompi-2022a
netCDF-Fortran/4.6.0-gompi-2022a
netCDF-Fortran/4.6.0-iimpi-2022a
netCDF-Fortran/4.6.0-iompi-2022a
OpenMPI/4.1.4-GCC-11.3.0
ParaView/5.10.1-foss-2022a-mpi
PnetCDF/1.12.3-gompi-2022a
PnetCDF/1.12.3-iimpi-2022a
PnetCDF/1.12.3-iompi-2022a
Python/3.10.4-GCCcore-11.3.0

---

[oyvindseland]

nco

[jmaerz]

Many thanks for opening this collection!

Some of the listed below are already in a newer version installed, which likely survive the upgrade.

• evince (light-weight pdf viewer)
• gitk (GUI for git branches, logs, etc.)
• ncl (?) - probably doesn't make much sense, while some older often used scripts still use it
• matlab
• netcdf4
• hdf
• openmpi
• paraview
• how about python2.7 - for older model/cime versions?
• pytorch

Not sure, if it should be only available via conda:

• ncl-, nco- and cdo-python bindings (while I saw from recent experiences that even their conda installs rely on availability of certain -GCC- libraries)

Maybe related: isn't there a possibility to declare defaults to load? - i.e. whenever wanting to load something, one seem to always has to search for a particular version, while it would simplify the module usage, when defaults would be predefined by sigma2 (but not necessarily loaded initially). Further, I never understood, when they use capital versus small letters in their naming convention...

[TomasTorsvik]

Most modules have a default version, marked with (D) if you look at modules avail, which is usually set to the latest version available. The problem with the defaults is that they may have different dependencies (e.g. try loading ncview and CDO modules on login-3.betzy, forces re-loading lots of modules with version changes).

[jmaerz]

Thanks - this is new - it wasn't working as expected under the former login nodes (at least not in my case). I always had to look up and load a specific version which can produce a lot of hassle.

Thus far, the list of available modules under login-3.betzy looks really short...

[TomasTorsvik]

Agree, I don't know it this is because it's work in progress, or this is what Sigma2 plan to go with.

[jmaerz]

Makes your initiative here even more important! - Thanks again for it.

[monsieuralok]

ESMF
ParMetis
Cmake
PARALLELIO

[hgoelzer]

Thanks Tomas,
great initiative.

Mostly in line with your list.
For my command line that I load in .bash_profile I am looking for a basic and consistent set including:
git
Python
ncview
NCO
CDO

Best greetings
Heiko

[irismuz]

Thank you for this initiative.
NCL would be very much appreciated.

[TomasTorsvik]

I will write to Sigma2/NRIS support about this now.

[TomasTorsvik]

Answer from Sigma2/NRIS

There is one available Python2 EasyBuild easyconfig, Python-2.7.18-GCCcore-12.3.0.eb. You can use this to install locally e.g in your Project area, see https://documentation.sigma2.no/software/userinstallsw/easybuild.html

NCL does not compile with the newer packages under the new OS. Maybe you can try the 'conda' solution https://www.ncl.ucar.edu/Download/conda.shtml? We will come back to you regarding Miniconda.