Running TauREx on the UCL HPC facilities (Legion/Grace/Myriad/Thomas/Aristotle) is slightly different than running on cobweb.

Fortunately, all UCL HPC systems work in the same way, so you can easily transfer your setup between, say, Legion and Grace. They have extensive documentation and user examples here. We will only discuss Legion and Grace here. There are differences between Legion, Myriad and Grace. Legion is a very heterogeneous system allowing small (<10 core) jobs, whilst Grace has a minimum requirement of 32 cores per job. For a breakdown of the specifications, see here.


Legion has a home and Scratch directory. During runtime, you must be in Scratch as home becomes read-only. So in Scratch, set up your TauREx folder.


This is ever changing. What works for now are the following combinations of modules:

module unload compilers/intel/2018/update3
module unload mpi/intel/2018/update3/intel

module load compilers/gnu/4.9.2
module load mpi/openmpi/1.10.1/gnu-4.9.2

module load python3/3.6
module load mpi4py/2.0.0/python3

Note that you must unload the default intel libraries to get mpi4py loaded. Best add these to your .bashrc.

Installing MultiNest & PyMultiNest

After loading the above modules, installing MultiNest is relatively easy on Legion. Follow the instructions here and make sure you export your library path (best add to .bashrc as well):


Finally, you must also install pymultinest. This is easily done with

pip install --user pymultinest

Submit Script

The submit script works differently to cobweb. Here’s an example:

#! /bin/bash -l
#$ -S /bin/bash
##name of job in queue
#$ -N [JOBNAME]       
#$ -V
##maximum walltime
#$ -l h_rt=5:00:00   
##maximum memory  
#$ -l mem=10G    
##maximum disk size      
#$ -l tmpfs=10G  
##number of cores      
#$ -pe mpi 16   
##path to working directory       
#$ -wd [PATH]          

##environment variables
export MPLBACKEND="pdf"

gerun python taurex.py -p [PARAMETERFILE] --plot

Note that gerun replaces all instances of mpirun on Legion. You can now submit and monitor progress in much the same way as on cobweb.