Skip to content

Slurm Fairshare User Manual

Overview

Fairshare (specifically, Fair-Tree Fairshare) scheduling in Slurm is a mechanism that allows equitable resource allocation among users and accounts based on their historical resource usage. This ensures all users receive a fair opportunity to access cluster resources, promoting a balanced utilization of GPU computational power.

Key Concepts

Accounts

  • Accounts: In Slurm, accounts are used to group users for resource allocation. Each account can have associated priorities and resource limits. In regards to Beatrix, all NRC users are equally prioritized for resource allocation.

  • Fairshare Weighting: Each account is assigned a fairshare factor that influences job prioritization. The factor is determined based on the account's historical usage and current resource limits. For more details on the specifics please see the slurm fair tree website

Partitions

  • Partitions: These are logical divisions within the cluster, allowing different sets of resources to be allocated for different types of jobs. Partitions can have different configurations, such as node types and resource limits.

  • Here are the partitions available on the Beatrix cluster

    Partition Time limit (D-HH:MM:SS) Nodes Note
    Trixiemain* 12:00:00 34 Default partition for all users
    Trixielong 2-00:00:00 24 Users must be authorized by the user representative
    JobTesting 6:00:00 2 Job testing partition
    Preemptible 12:00:00 30 Jobs submitted here may be pre-empted
    Larus 7-00:00:00 34 Industry SME partition, not available to NRC users

Fair-Tree Fairshare Scheduling

How It Works

  • Historical Usage: Slurm tracks the historical usage of resources by each account, adjusting priorities based on this data. Accounts using fewer resources within the fairshare window will have a higher priority relative to other users who have used more resources. This calculation is done at multiple levels, such that if accounts A and B are siblings and A has a higher fairshare factor than B, all children of A will have higher fairshare factors than all children of B.

  • Priority Calculation: Job priorities are calculated using a combination of fair-share factors, job age, and other configurable attributes.

  • Dynamic Adjustment: The system dynamically adjusts priorities, ensuring fair distribution of resources over time.