Create a resource quota - Platform For AI - Alibaba Cloud Documentation Center

After purchasing Intelligent Computing Lingjun or general computing resources, create a resource quota for hierarchical management and isolated use of your resources. With resource quotas, you can control the resource usage limits for different teams or projects and bind them to workspaces for AI development and inference.

Prerequisites

Before you create a resource quota, complete the following prerequisites:

You have purchased Intelligent Computing Lingjun resources or general computing resources.
(For Intelligent Computing Lingjun resource quotas only) You have created a VPC, a vSwitch, and a security group.

Add a resource quota

PAI supports two types of resource quotas: Intelligent Computing Lingjun resource quotas, which require network configuration (VPC, vSwitch, and security group), and general computing resource quotas, which do not.

When you create a resource quota, you can either allocate resources from a dedicated resource group in your resource pool to create a root quota, or from an existing resource quota to create a child quota.

Important

For Intelligent Computing Lingjun resources, all nodes must use the same hz identifier to ensure high-speed network connectivity between them.

Follow these steps:

Log on to the PAI console and go to the AI Computing Resources > Resource Quota page.
On the Lingjun Intelligent Computing resources or General Computing Resources tab, click Add Resource Quota.
Configure basic information:
- Name: Enter a name for the resource quota.
- Associate Workspace (Optional): Select a workspace to associate with this quota, which allows you to use it in that workspace.
  
  Note
  If you do not associate a workspace now, you can bind one later on the resource quota details page.
Configure quota resource information:
1. Select a Source Type. The following types are supported:
  - Dedicated Resource Group: Select a resource group from your resource pool. The system allocates resources from this group to create a root quota.
  - Existing Resource Quota: Allocate resources from an existing resource quota to create a child quota.
  For more information about the relationship between parent and child quotas, see Resource Quota (Quota).
2. From the drop-down list, select an existing dedicated resource group or resource quota.
3. Click Add and select a Nodes/Instance Type from the existing resource quota or dedicated resource group.
Configure scheduling information:
- Scheduling Policy: Select a scheduling policy to improve compute resource utilization. The following policies are supported:
  - Intelligent
  - Balance
  - Traversal
  - FIFO
  For information about how each scheduling policy works, see Scheduling Policy.
- Child-level Preemption (Optional): When enabled and resources are scarce, queued tasks in the current resource quota can preempt running tasks in its child quotas. For more information, see Enable child-level preemption.
- Self-level Preemption (Optional): When enabled and resources are scarce, queued tasks in the current resource quota can preempt running tasks in other quotas at the same level. For more information, see Enable same-level preemption.
- Idle Sharing: Enabled by default. This allows idle tasks to use available resources from quotas at the same level and in child quotas.
(For Intelligent Computing Lingjun resource quotas only) Configure network information:

Configure network settings to ensure instances within the resource quota can access specified VPC network resources.
- VPC: Select an existing VPC.
- Security Group: Select an existing security group.
- vSwitch: Select an existing vSwitch.
- Default Internet Gateway (Optional): If you need to access the internet, enable this switch and select a NAT Gateway and an EIP.
Click OK.

After you create the resource quota, it appears in the resource quota list.

Use a resource quota

After creating a resource quota, you need to bind it to a workspace and then select a suitable image for AI development and inference.

Bind to a workspace

You must bind a resource quota to a workspace before you can use it for AI development and inference. Follow these steps:

Note

You can skip this step if you already associated a workspace when you created the resource quota.

On the Resource Quota (Quota) page, click the name of the resource quota.
On the Overview tab, in the Basic Information section, click the icon next to Workspace to add or modify the associated workspace.

After you bind the resource quota to a workspace, you can also configure its usage policies on the Scheduling Configuration page within the workspace details. For more information, see Workspace Scheduling Center.

Image selection

When you use a PAI Lingjun resource quota for distributed training (DLC), it involves the coordination of software and hardware, such as servers, the network, drivers, and training frameworks. Therefore, we recommend that you use official PAI images or build images based on them.