Technical Bulletin #518: Moab to Slurm Replacement Schedule
The license to run the Moab Workload Manager (Moab commands) will expire this year on September 30 and will not be renewed. The Tri-Labs have elected to activate SLURM’s scheduling plugin to provide the functionality formerly provided by Moab. Machines running TOSS 3 are already running SLURM scheduling.
Over the next few months, the remaining clusters that are still scheduled by Moab will be converted to SLURM scheduling (see table below, in order of retirement). Other systems using Moab are herd, rzzeus, rzmerl, but all of these systems are scheduled to retire over the next month or so, thus they will run Moab until they retire.
During each conversion, all jobs running on the machine will be killed and all jobs queued to the machine being converted will be cancelled. Once the machine is reconfigured, queueing will begin again. In order to smooth the transition to SLURM, Livermore Computing is providing wrappers that will allow most of the Moab commands (e.g., msub) to work as they currently do. In addition, users can run the moab2slurm script to convert any of their Moab scripts (#MSUB directives) to SLURM equivalents (#SBATCH directives).
After the conversion, users are encouraged to start using the native Slurm commands as there are Slurm command options available that have no Moab equivalents.
Native SLURM commands can be found here: https://hpc.llnl.gov/banks-jobs/running-jobs/slurm-commands
More detailed SLURM info:
Printable pdf: PDF of TB518 for distribution.