Setting Up License Checking for Queueing Systems
On large clusters, it is possible for jobs to fail because there are too few licenses for all the jobs to run. A mechanism has been provided to ensure that a queued job will not run unless all the licenses it requires are available. This facility is available for:
- Grid Engine queuing systems (SGE)
- PBS Pro queueing system from version 10.1 on
- Platform LSF queuing system from version 6.2 on
- Slurm version 18.08.2 or newer
SGE, PBS Pro, and LSF
The configuration process for SGE, PBS Pro, and LSF consists of the following steps:
- Define a resource representing each license in the queuing system configuration.
- Configure the load sensor script, which reports how many tokens are available for each license, so that it can find the license file and the FLEXlm utility that it uses to query the license server. It can also be configured to reserve licenses for non-queue use.
- Install this script in the appropriate queuing system directory.
- Configure Job Server to specify license requirements when launching jobs.
Each of these steps is described in detail in the topics listed below.
- Adding License Attributes
- Configuring the Load Scripts
- Installing the Load Script
- Turning On License Checking in Job Server
Slurm
The Slurm queueing system offers two different methods of license checking, “local” and “remote” (https://slurm.schedmd.com/licenses.html). Local license checking is easier to implement but is only applicable when the license is used by jobs running on a single Slurm cluster. Remote license checking requires more configuration but can cope with multiple clusters and license checkouts from systems outside a Slurm cluster, e.g., laptops and workstations.
Follow these steps for local license checking:
- Configuring Local License Checking for Slurm
- Turning On License Checking in Job Server
- Testing License Checking for Slurm
Follow these steps for remote license checking: