SageMaker Studio Enterprise Deployment guidelines


Do we have guidelines on requirements gathering/designing the provisioning of SageMaker Studio domains across large global enterprises with many business units?

I've seen discussions where topics like number of users/domain, org/team structure, collaboration patterns, resource needs, classes of ML problems, framework/library usage, security and others were raised when defining requirements and boundaries. Customer is starting their first Studio deployment and they are asking for guidance on how to scope and design that so that they can have a scalable process.

質問済み 3年前379ビュー

You should guide your customer based on the general principles of multi account best practices that we provide for other services.

Here are some high level boundaries.

One studio domain per account and region. No cross region AWS SSO configuration provided.

Maximum numbers of users allowed in studio vary between 60 - 200 users. Although AWS SSO can support many more users, there are some considerations around other dependencies such as EFS among others.

If you need to isolate any model artifacts produced by SageMaker, you may want to have them use a separate account. Even if you use tag based access control, you can still technically list those artifacts.

SageMaker feature store should follow the data lake pattern closely. As a general rule, you want to write in one account and can read from many other accounts perhaps using Lake formation to expose datasets into other accounts. Teams can create their own offline / online feature store for non production use cases.

回答済み 3年前

ログインしていません。 ログイン 回答を投稿する。