Slurm

Slurm is an open-source workload manager designed for Linux computer clusters, used to allocate resources, schedule jobs, and monitor cluster usage in high-performance computing environments. It emphasizes scalability, reliability, and a modular architecture suitable for large-scale systems.

The Slurm architecture centers on a controller and daemons: a central controller process (slurmctld) runs on the management node, while compute nodes run the slurmd daemon. An optional accounting daemon (slurmdbd) can store usage data in a relational database. Configuration is provided by the slurm.conf file, which defines partitions, nodes, and scheduling policies. Authentication typically uses Munge; LDAP and other methods can be integrated as needed.

Scheduling and resource management features include partitions, which group nodes for different workloads, and information about nodes such as state and features. Jobs are submitted with sbatch or srun and can be organized into job arrays, have dependencies, or be restricted by quality of service (QoS) and reservations. Slurm supports backfill scheduling and, in some configurations, preemption to improve resource utilization and meet priority requirements.

Accounting and monitoring capabilities enable tracking of usage and performance. Slurm can store accounting data in a relational database via SlurmDBD, with common backends including MySQL, MariaDB, or PostgreSQL. User-facing and administration tools include sacct, sacctmgr, sreport, squeue, scontrol, and sprio, which provide status, accounting, and priority information as well as cluster control.

History and licensing: Slurm originated at Lawrence Livermore National Laboratory in the 1990s as Simple Linux Utility for Resource Management and was later renamed Slurm. It is released under the GNU General Public License, with community and vendor support continuing to evolve. Slurm is widely deployed in research and industry HPC environments and maintained by a broad developer community.