You can purchase and create different types of MaxCompute resource quotas, associate the quotas with Platform for AI (PAI) to develop models, and then submit related jobs to MaxCompute to implement large-scale distributed training. This topic describes MaxCompute resources and provides instructions on how to configure the resources.
MaxCompute resources
MaxCompute is an enterprise-level cloud data warehouse that uses the software as a service (SaaS) model. MaxCompute is suitable for scenarios that involve data analytics. It provides a fast, fully-managed online data warehousing service in a serverless architecture. MaxCompute eliminates the constraints of traditional data platforms in terms of resource extensibility and elasticity, minimizes O&M costs, and allows you to efficiently process and analyze large amounts of data at low costs. For more information, see What is MaxCompute?.
Types of MaxCompute resources
The following section describes the types of MaxCompute resources supported by PAI.
CPU resources
In PAI, CPU resources are billed in the same way as in MaxCompute. The following resources are supported:
Standard pay-as-you-go resources (recommended)
Standard subscription resources
GPU resources
GPU resources are available only in PAI. The following types of resources are supported:
Standard pay-as-you-go resource (recommended)
Standard subscription resource
The following table describes each type of resources.
You are charged for the following categories of MaxCompute resources: computing resources, storage resources, and resources for data uploads and downloads. In most cases, resources for data uploads and downloads are not required for training jobs in PAI.
If you consume pay-as-you-go computing resources, fees incurred by computing jobs are charged in PAI whereas fees incurred by other jobs, such as SQL jobs, are charged in MaxCompute.
Item | Standard pay-as-you-go resource (recommended) | Standard subscription resource |
Computing resources (used in training jobs) | The resources come from a shared resource pool. If you use this resource category, computing jobs in your MaxCompute project can preempt resources in the resource pool when necessary. | The resources contain exclusive and shared computing resources. Shared computing resources are optional. |
Storage resources (used to store training result tables) | The resources come from a shared resource pool. You can use storage resources in the resource pool on demand. | The resources come from a shared resource pool. You can use storage resources in the resource pool on demand. |
Billing rules for computing resources | ||
Billing rules for storage resources | N/A | |
Scenarios | Projects that require flexible storage space. | Projects that have been created and are running properly for your business. |
Account and permission requirements
Alibaba Cloud account (recommended): You can use an Alibaba Cloud account to complete all operations without additional authorization.
RAM user:
If you want to use a RAM user to purchase MaxCompute resources, you need to grant the
AliyunBSSOrderAccessandAliyunDataWorksFullAccesspermissions to the RAM user. For more information, see Prepare a RAM user.If you want to use a RAM user to associate MaxCompute resources with a workspace, you need to assign the administrator role to the RAM user in the workspace. For more information, see Manage the members of a workspace.
If you want to use MaxCompute resources to train models in Machine Learning Designer, you need to assign the MaxCompute developer role to the RAM user in the workspace. For more information, see Manage the members of a workspace.
Activate MaxCompute and purchase MaxCompute resources
You can go to the MaxCompute buy page to purchase MaxCompute resources based on your needs.
Use and manage MaxCompute resources
After you associate a purchased MaxCompute resource group with a PAI workspace, you can deploy jobs to the group in the workspace. You can also manage the resource groups on the Pay-As-You-Go Quota Groups page of the MaxCompute console.
Associate a MaxCompute resource group with a workspace
When you create a workspace, you can associate a MaxCompute resource quota with the workspace. When you associate a MaxCompute resource quota with a workspace, you can edit the name of the project and specify whether to enable GPU. GPU can be enabled only in specific regions. You can use the associated MaxCompute resources for all your training jobs created in the workspace.
NoteEach workspace can be associated with only a single MaxCompute project.
You can also modify the MaxCompute project associated with a workspace after the workspace is created. For more information about how to create a workspace, see Create a workspace.
Manage MaxCompute resources
You can log on to the PAI console and choose to view the available MaxCompute resources in the current region.
You can view the total MaxCompute CU quotas and the billing methods of MaxCompute resources.
You can click Resources to go to the MaxCompute console to manage quotas for pay-as-you-go resources. For more information about how to manage quotas, see the "Configure quota groups" section in the Manage quotas in the new MaxCompute console topic.