Skip to content

Resource Allocation

Tenant admins can allocate computing resources (GPU, CPU, memory) to specific workspaces and Kubernetes namespaces. Use this tool to ensure efficient distribution of resources within your tenant.

Resource Summary

1. Resource Distribution Overview

A top section summarizes total, allocated, and available resources:

  • GPUs
  • Memory
  • CPUs

This snapshot helps you quickly gauge remaining capacity and usage levels.

2. Resource Allocation Interface

Two tabs let you manage resources for:

  1. Workspaces
  2. Namespaces

2.1 Workspaces Tab

View each workspace’s resource usage with:

  • Workspace Name
  • Allocated GPUs
  • Allocated Memory
  • Allocated CPUs
  • Actions (Allocate Button)

Allocating Resources to Workspaces

  • Click Allocate next to a workspace.
  • Set GPU, memory, and CPU limits in the popup.
  • Save to apply changes.

Workspace Allocation Popup

3. Resource Sharing

  • When you assign resources to a workspace, it is guaranteed that the workspace will receive the allocated amount (essentially increasing the workspace's priority to take resources from the shared pool)
  • If a workspace or namespace isn’t explicitly assigned resources, it defaults to shared usage of the tenant’s overall capacity. This ensures all entities have access to essential resources, subject to availability.
  • Additionally, reclaim feature of Volcano is supported. Reclaim is one of the actions in kube-batch scheduler. Reclaim action comes into play when a new queue is created, and new job comes under that queue but there is no resource / less resource in cluster because of change of deserved share for previous present queues. Reclaim is evicting tasks from other queues so that present queue can make use of it's entire deserved share for creating tasks.

Next Steps