Create a resource quota-Platform For AI(PAI)-阿里云帮助中心

After purchasing Lingjun intelligent computing resources or general-purpose computing resources, you must create resource quotas to manage and isolate resources in a hierarchical structure. Resource quotas control the resource usage limits for different teams or projects. After a quota is bound to a workspace, the workspace can use it for AI development and service inference.

Prerequisites

Before creating a resource quota, you must purchase Lingjun intelligent computing resources or general-purpose computing resources.

Add a resource quota

PAI supports two types of resource quotas: Lingjun intelligent computing resource quotas, which require VPC, vSwitch, and security group configuration, and general-purpose computing resource quotas, which require no network configuration.

When creating a resource quota, you can allocate resources from a dedicated resource group in the resource pool to create a root-level quota, or allocate resources from an existing quota to create a child quota.

Important

Lingjun intelligent computing resources: All nodes must use the same Hz value to ensure high-speed network interconnection between nodes.

Log on to the PAI console and navigate to the AI Computing Resources > Resource Quotas page.
On the Lingjun Intelligent Computing resources or General Computing Resources tab, click Add Resource Quota.

Configure the basic information:

Parameter	Description
Name	Enter a name for the resource quota.
Associate Workspace	Select the workspace to bind the resource quota to. After binding, the quota can only be used within that workspace. Note Each quota can be bound to only one workspace. Cross-workspace sharing is not supported. You can skip binding for now and bind the quota later from the quota details page.
Tag	Add tags to instances for multi-dimensional search, filtering, batch operations, and cost allocation.

Configure the quota resource information:

Parameter	Description
Source Type	Dedicated Resource Group: Allocate resources from a dedicated resource group in the resource pool to create a root-level quota. Existing Resource Quota: Allocate resources from an existing quota to create a child quota. For more information about the relationship between parent and child quotas, see Resource quotas overview.
Source	Select an existing quota or a dedicated resource group.
Nodes/Instance Type	Click Add and select a node or specification from the existing quota or dedicated resource group.

Configure the scheduling information:

Parameter	Description
Scheduling Policy	Select a scheduling strategy to improve resource utilization. The supported strategies are Smart, Balanced, Round-robin, and FIFO. For more information, see Scheduling policies.
Child-level Preemption	When enabled, queued jobs in the current quota can preempt running jobs in child quotas when resources are scarce. For more information, see Enable child-level preemption.
Self-level Preemption	When enabled, queued jobs in the current quota can preempt running jobs in sibling quotas when resources are scarce. For more information, see Enable sibling-level preemption.
Idle Sharing	Enabled by default. When enabled, idle resources from sibling and child quotas can be borrowed by jobs in this quota.

Configure the network information (required only for Lingjun intelligent computing resource quotas) to allow instances in the quota to access the specified VPC network resources:

Parameter	Description
VPC	Select an existing VPC.
Security Group	Select an existing security group.
vSwitch	Select an existing vSwitch.

Click OK.

After the quota is created, it appears in the resource quota list.

Use a resource quota

Bind a workspace

A resource quota must be bound to a workspace before it can be used for AI development and service inference within that workspace. To bind a quota to a workspace:

On the Resource Quota page, click the name of the target quota.
In the Overview tab, locate the Basic Information section, and click next to Workspace to add or modify the workspace binding.

You can also bind a resource quota from the workspace side to achieve the same result:

Log on to the PAI console, navigate to the details page of the target workspace, and click Workspace Settings > Compute Resource Settings.
Select the Lingjun Intelligent Computing resources or General Computing Resources tab. In the Available Resources list, select the quota and click Confirm Association.

Note

The quota and workspace must be in the same region. A single workspace can be associated with multiple quotas.

After binding, you can configure quota usage policies in the Scheduling Settings of the workspace details page. For more information, see Workspace scheduling center.

On the Configure Workspace page, click the Scheduling Settings tab. In the Resource Usage section, you can view and modify the following settings: Resource Usage Module, Resource Usage Role, Resource Usage Quota, and Resource Specification Template.

Access the public network

To enable public network access — for example, to pull custom images or download third-party packages and models — configure the SNAT feature of a NAT gateway and bind an elastic IP address (EIP). For more information, see Use the SNAT feature of an Internet NAT gateway to access the Internet.

Select an image

When using Lingjun intelligent computing resource quotas for distributed training (DLC), coordination across servers, networks, drivers, and training frameworks is required. We recommend using official PAI images or building custom images based on official images.