Skip to content

Deployment Options

Two managed SLURM on Kubernetes options are available.

Standard Edition

With this approach, multiple SLURM clusters share a common Kubernetes host cluster. Each SLURM cluster operates in owns namespace and will have hard limits for resource quotas to prevent noisy neighbor issues. This is ideal for quick deployments with smaller workloads.

Characteristics

  • User Experience: SLURM clusters can be deployed for users in < 5mins.
  • Scalability: Easy to add or remove SLURM clusters with minimal infrastructure churn.
  • Resource Pooling: Nodes, GPUs, and storage can be dynamically allocated among SLURM clusters.
  • Efficiency: Shared control plane and system services reduce management overhead.

Use Cases

  • Multi-tenant AI or GPU environments where performance isolation is sufficient at the namespace level
  • Centralized infrastructure teams managing large shared compute environments.
  • Cost-sensitive deployments that benefit from shared management and elastic scaling.

Multi Tenant Slurm

Pro Edition

With this approach, each SLURM cluster is mapped to its own dedicated Kubernetes cluster. This approach provides strong isolation between workloads and tenants, ensuring that SLURM and Kubernetes resources do not interfere across environments.

Info

This is a custom, expert-run deployment designed for large and complex workloads. Our experts will assist with both the configuration and deployment of the infrstructure, SLURM cluster with your list of custom applications.

Characteristics

  • Isolation: Each SLURM cluster operates independently within its own dedicated Kubernetes cluster
  • Security: Tenant-level separation simplifies compliance and minimizes attack surfaces.
  • Customization: Each Kubernetes cluster can be tuned for specific workloads (e.g., GPU-heavy, AI/ML, or HPC).
  • Resource Utilization: Typically lower, as resources cannot be easily shared across clusters.

Use Cases

  • High-security or regulated environments requiring full workload isolation.
  • Research institutions or enterprise divisions with distinct operational requirements.
  • Scenarios where per-tenant performance tuning or custom networking is essential.

Multi Tenant Slurm