Configuring workload resource profiles

Each project editor session and deployment consumes compute resources within the Workbench cluster. If you need to run applications that require more memory or compute power than Anaconda provides by default, you can create customized resource profiles and configure the number of cores and amount of memory/RAM available for them. You can then allow users to access your customized resource profiles either globally or based on their role, assigned group, or as individual users. For more information about groups and roles in Workbench, see Roles. For example, if your installation includes nodes with GPUs, you can add a GPU resource profile so users can access the GPUs to accelerate computation within their projects, which is essential for AI/ML model training. Resource profiles that you create are listed for users when they create or deploy a project. You can create as many resource profiles as needed.

Configuring resource profiles via Helm chart

Workbench includes a values.yaml file that overrides the default values in the top-level Helm chart. If you are performing your initial configurations, use the examples below to add the necessary resource-profiles: and gpu-profile: sections to the bottom of your file, then continue your environment preparation and installation. Otherwise, follow the steps for Setting platform configurations using the Helm chart to add or update the resource-profiles: and gpu-profile: configurations to the bottom of your values.yaml override file as needed for your system setup.

Resource profile examples

resource-profiles:
  resource-profile:
    description: Custom resource profile (global)
    resources:
      limits:
        cpu: "2"
        memory: 4096Mi
    user_visible: true

  roles_profile:
    description: Custom resource profile (roles)
    resources:
      limits:
        cpu: "3"
        memory: 4096Mi
    user_visible: true
    acl:
      roles:
        - ae-creator

  groups_profile:
    description: Custom resource profile (group)
    resources:
      limits:
        cpu: "4"
        memory: 4096Mi
    user_visible: true
    acl:
      groups:
        - managers

  users_profile:
    description: Custom resource profile (users)
    resources:
      limits:
        cpu: "1"
        memory: 4096Mi
    user_visible: true
    acl:
      users:
        - user2

  not_visible_to_anyone:
    description: Custom resource profile (nv)
    resources:
      limits:
        cpu: "3"
        memory: 4096Mi
    user_visible: false

  gpu-profile:
    description: GPU resource profile
    resources:
      limits:
        cpu: "4"
        memory: 8Gi
        nvidia.com/gpu: 1
      requests:
        cpu: "1"
        memory: 2048Mi
        nvidia.com/gpu: 1
    user_visible: true

Resource profiles display their description: as their name. Profiles are listed in alphabetical order, after the default profile.

By default, CPU sessions and deployments are allowed to run on GPU nodes. To reserve your GPU nodes for sessions and deployments that require them, comment out the affinity: configuration in the file as shown:

If you need to schedule user workloads on a specific node, add a node_selector to your resource profile. Use node selectors when running different CPU types, such as Intel and AMD; or different GPU types, such as Tesla v100 and p100. To enable a node selector, add node_selector to the bottom of your resource profile, with the model: value matching the label you have applied to your worker node.

GPU node selector example

gpu-profile:
  description: GPU resource profile
  resources:
    limits:
      cpu: "4"
      memory: 8Gi
      nvidia.com/gpu: 1
    requests:
      cpu: "1"
      memory: 2048Mi
      nvidia.com/gpu: 1
  user_visible: true
  node_selector:
    model: v100

Once you have run the Helm upgrade command, verify that the resource profiles you added appear in the Resource Profile dropdown under your project’s Settings.

Data Science & AI Workbench (5.8.0)

​Configuring resource profiles via Helm chart

Configuring resource profiles via Helm chart