Looking to upgrade your Bright Cluster Manager to 8.0? Depending on your existing cluster components, you will need to update your OS and scheduler as well. Here is a guideline on how to deploy a Bright/RHEL7/Slurm upgrade.
- Prepare environment for a Bright/RHEL7/Slurm upgrade.
- Inspect system state, queues and health. Backup OS, applications and settings.
- Prepare environment for a parallel Bright environment on a virtual one using Vsphere.
- Create new RHEL7 Head Node
- Install Bright to v 8.0
- Install new virtual image with Bright to version 8.x
- Install Slurm version 17.x
- Develop functional compute, KNL and GPU CentOS 7.x image
- Deploy to at least one test node.
- Buildout an application tree that is interoperable with CentOS 7.
- Test Slurm queue with test node using applications on a CentOs 7.x image.
- Push out through Bright RHEL 7.x images to all compute nodes
Ideally, existing SLURMDBD accounting databases will be maintained for past job data and existing MYSQL databases for XDMod will be maintained. This however, is dependent on the SLURMDBD configuration for new jobs. It is possible to ingest old data so it may be easier to maintain SLURMDBD databases and reimport.
Other software may need to be implemented and configured in the new image. For example, Latest OpenMPI will be needed to be compliled against Intel Complier. Applications like Lammps, Gaussian, and VASP will need to be compiled and may need separate versions for optimization against CUDA, KNL and CPU.
Need more insight? We can take your Bright upgrade project off your hands! Contact DST at firstname.lastname@example.org.