Platform for AI (PAI) provides a resource management and scheduling mechanism that allows workspace administrators to schedule resources in a workspace based on business requirements.
Prerequisites
A workspace is created. For more information, see Create a workspace.
Configure a policy
Log on to the PAI console.
In the left-side navigation pane, click Workspaces. On the Workspaces page, click the name of the workspace that you want to manage.
On the Workspace Details page, click the Scheduling Center tab.
Find the policy that you want to modify and click Modify Configuration next to the policy. The following table describes the details of the policy.
Policy
Description
General Configuration Policy
Modules
Resource Quota: Select the resource quotas that are associated with the workspace.
Supported Modules: Valid values: DSW and DLC.
Roles
Resource Quota: Select the resource quotas that are associated with the workspace.
Supported Role: Valid values: Basic Role (Administrator, Algorithm Developer, and Algorithm O&M Engineer. ), Custom role, and Non-workspace members.
NoteNon-workspace members are not members of the workspace, but are granted the related RAM permissions by the Alibaba Cloud Account. Non-workspace members can use resources and submit tasks. You can make custom control policy for them.
GPUs
Resource Quota: Select the resource quotas that are associated with the workspace.
Role: Valid values: Basic Role (Administrator, Algorithm Developer, and Algorithm O&M Engineer. ), Custom role, and Non-workspace members.
Maximum Number of GPUs: Specify the maximum number of GPUs.
Resource Specification Templates
Resource Quota: Select the resource quotas that are associated with the workspace.
Template Nmae: Specify a name for the resource specification template.
vCPUs: The number of vCPUs that the template supports.
Memory (GiB): The memory usage that the template supports.
GPUs: The number of GPUs that the template supports.
Supported Modules: Valid values: DSW and DLC.
DLC Policy Configuration
Maximum Running Duration
You can specify a maximum running duration for Deep Learning Containers (DLC) jobs in the current workspace by day, hour, or minute.
Job Priority
Type: Valid values: Workspace Role and Workspace Member.
Scope:
Valid values if you set the Type parameter to Workspace Role: Basic Role (Administrator, Algorithm Developer, and Algorithm O&M Engineer. ), Custom role, and Non-workspace members.
Valid values if you set the Type parameter to Workspace Member: Alibaba Cloud accounts.
Highest Priority: The priority ranges from 1 to 9. A greater value indicates a higher priority.
DSW Policy Configuration
Automatic Recycling
Click Add and configure the Hour parameter. The Data Science Workshop (DSW) instances in the current workspace are automatically stopped after running for the specified hours.
Click Add Group and configure the Idle Duration, CPU Utilization, Memory Utilization, or GPU Utilization parameters. The DSW instances in the current workspace are automatically stopped if the specified conditions are met.
NoteIn a policy group (AND), DSW instances are automatically recycled if all conditions are met.
In all policy groups (OR), DSW instances are automatically recycled if one of the conditions is met.
Click Save.
Related operations
On the Scheduling Center tab, you can also perform the following operations:
Create a policy
If the current policies do not meet your business requirements, you can click Add to create policies.
Delete a policy
The following section provides an example of deleting a policy in the Used Modules section.
1. Click Delete to delete all policies in the Used Modules section.
2. Click Delete to delete a single policy.
