Purdue University

As of 4:00 PM on Thursday, August 6th, 2020, the Software Stack Transition on Brown, Gilbreth, Halstead, Rice, Snyder, and Workbench clusters have been completed. If you have a login session (Thinlinc, Gateway, SSH, etc.) that was started before the transition, please log out and start a new session. If you continue using the old session, you may get inconsistent set of modules.

More details about the new software stack and answers to frequently asked questions may be found in the FAQ page.

---

To expand the list of supported software on ITaP Community Clusters and to provide software updates at a faster pace, Research Computing is adopting a new software automation framework (Spack) that will replace our existing software stack. Applications built with Spack have been deployed on Brown, Gilbreth, Halstead, Rice, Snyder, and Workbench clusters and are now available for testing via the module command. These modules will replace existing modules on Thursday, August 6th, 2020, therefore, users are requested to follow the instructions below to test their workflows with the new modules.

We have tried to keep the list of applications and their respective versions similar to the old applications. However, in some cases, the application names and minor versions may have changed. If you find that your code stopped working with the new modules, please find the correct applications and versions using the module spider command and update your scripts.

What you need to do

Please test your jobscripts and your workflow with the new modules as soon as possible. This will help avoid surprises when we make a complete transition on Thursday, August 6th, 2020.

To test the new modules

• First load the modtree/new module: module load modtree/new.
• Now you can use the module commands as before, e.g, module spider openblas.

$ module load modtree/new
$ module spider openblas
$ module load openblas/0.3.7
... ... ...

To go back to the old modules

• Load the modtree/deprecated module: module load modtree/deprecated.
• Now you can use the old modules as before, e.g., module spider openblas.

$ module load modtree/deprecated
$ module spider openblas
$ module load openblas/0.2.20
... ... ...

Avoiding job failures during transition

Jobs that are submitted prior to Thursday, August 6th, 2020, but do not start before the transition, may fail when the modules are replaced. To avoid job failures, please include the following line (once) before all module load commands in your jobscript.

module load modtree/new

Make sure that you have tested your jobscripts with this addition.

Highlights

• The new software stack will resemble the modules already in production on the Scholar cluster. So, if your jobscript works on Scholar, it is expected to work on other clusters after the transition.
• Some applications and libraries (HDF5, NetCDF, CDO, NCO, ncview, etc.) which were previously combined under a single umbrella module (netcdf) have been separated into individual modules.
• Embedded gmp, mpc, and mpfr libraries have been separated from GCC into their own modules. They will be automatically loaded when you load the gcc module.
• The lapack module has been renamed as netlib-lapack. If your application requires LAPACK, you can also use the openblas module and link with -lopenblas flag.
• The pnetcdf module has been renamed as parallel-netcdf.
• The default Anaconda version has changed from 5.1.0-py27 to 5.1.0-py36.
• The default R and Rstudio versions have been upgraded to 3.6.3 and 1.3.959 respectively.
• Starting with R 3.6.3, we no longer embed external applications (such as HDF5, NetCDF, etc.) into the R installation. If you need to use R-packages that depend on these libraries, you must load the corresponding module(s) separately.
• You will find several new versions for other applications and libraries.

Bioinformatics

Due to the complexity of Bioinformatics software collection and its dependency on old RCAC software, these applications will NOT use the new software stack. When you load the bioinfo module, your environment will be automatically configured to use bioinformatics packages and their dependency modules. You may notice duplicate messages while loading the bioinfo module -- this is expected.