Important Notes from the 17 September Rivanna Maintenance

Rivanna was down for maintenance on Tuesday, September 17. The items below summarize the changes that may impact the users of Rivanna.

I. Changes to scratch

System engineers have installed a new /scratch file system, and have transferred to the new system any files/data that were less than 90 days old on the former scratch system.

II. Updates to software modules

New and updated modules:

The following software modules either replace older versions or are new to Rivanna:

 - pgi/19.7
 - openmpi/3.1.4 (for all GCC and PGI compilers)
 - cuda/10.1.168

For openmpi, be sure to remove any reference to 2.1.5 in your scripts.

Removed modules:

The following software modules were removed from Rivanna during the maintenance period:

- cellranger/2.1.1 (replaced with cellranger/3.1.0)
- exonerate/2.2.0 (replaced with exonerate/2.4.0)
- fenics/20180
- fluent/18.2 (is now part of the ansys/18.2 module)
- fiji/1.51
- miniconda/4.3.21-py3.6 (replaced with anaconda/5.2.0-py3.6
- openmpi/2.1.5 (replaced with openmpi/3.1.4)
- pgi/17.5 &  pgi/18.10 (replaced with pgi/19.7)
- povray/
- rstudio/0.98.1103

III. Other important changes

The loading of some software modules now requires preloading of a dependency, such as a compiler or version of mpi.

Run the command module spider <YOUR_MODULE> to view module load instructions for a particular application module.

For example, module spider abinit/8.2.2 states that

 You will need to load all module(s) on any one of the lines below before the "abinit/8.2.2" module is available to load.
 intel/18.0  intelmpi/18.0

This statement tells you that both intel and intelmpi must be loaded in order to load abinit.

Recompiling & Reinstalling

The operating system was updated, and (as usual) users who compile their own code may need to recompile.

This also applies to anyone who installed R packages which are dependent on openMPI. Those packages will need to be reinstalled.

Libraries and applications built with the Intel 18.0 compiler and IntelMPI libraries have been re-compiled to enable execution on compute nodes with Knights Landing Many-Core processors in the knl queue.

If you have any questions or concerns about these changes, please contact our user support team.