How to install Savu for a HPC cluster

Savu is designed to run on a High Performance Computing cluster. For optimum performance, we recommend a fast network interconnect, such as infiniBand, and a high performance parallel filesystem, such as GPFS or Lustre.

Requirements:

  • A Unix system with MPI libraries installed: Savu has been tested with openMPI (version 4.1.1)

Installation of Savu HPC outside Diamond Light Source (DLS) systems

  1. Check that you do NOT have conda in your path with “which conda”. Note that the installer will install its own version of conda so it is essential that conda is not in the path.

  2. Download savu installer and extract the archive.

  3. Select the desired Github branch of Savu to install with export savu_branch=”master” or specify Savu version explicitly export savu_branch=”savu_version”.

  4. Set the name of the facility export facility=’facility_name’.

  5. >>> bash savu_installer/savu_installer.sh

  6. Check the log file /tmp/<tmpfolder>/savu_error_log.txt for installation errors (correct log file path printed to screen during installation process).

Installation of Savu HPC at Diamond Light Source (DLS) systems (fastest)

This is the fastest installation which uses pre-built against openmpi packages _mpi4py, hdf5, h5py_ from savu-dep conda channel.

  1. Do 1-3 steps as above.

  2. >>> export explicit_file='savu_list_openmpi4_1_1.txt'

  3. >>> bash savu_installer/savu_installer.sh

Installation of Savu HPC at Diamond Light Source (DLS) systems

When the explicit packages list file is not provided, the installer will use environment.yml file to install additional to pre-built packages. This will be slower than the previous option.

Installation for Diamond Light Source (DLS) systems:

  1. Get savu installer

  2. >>> bash savu_installer/savu_installer.sh