aws / aws/amazon-sagemaker-feedback

Faster Launch Times

Open
#189 0 comments 3 reactions 0 assignees View on GitHub
feature-request
Dominant language
No language data
Stars
10
Forks
3
PR merge metrics
No merged PRs in 30d

Description

### Product Version

- [ ] Amazon SageMaker Studio Classic
- [x] Amazon SageMaker Studio
- [ ] It is not related to SageMaker Studio

### Product Category

Other

### Description

My usual experience with SageMaker

'Creating ML flow server'
Note: This may take 20-25 minutes.

'Registering Model'
Note: This may take 20-25 minutes.

'Creating Endpoint'
Note: This may take 20-25 minutes.

etc. etc. etc.

EFS can spin up instances in seconds. Why is SageMaker so slow? Even CPU-only instances take 20-30 minutes to deploy.

It's not realistic to expect that we can scale from zero and wait 15-30+ minutes for even the simplest model endpoint.

Compare this complaint of a Fargate user:
https://repost.aws/questions/QUjZAzJd27SZWxXM7MgyxOZw/how-to-speed-up-provisioning-of-ecs-fargate-task (complaining about the long 30-40 SECONDS of delay 😂. I can only wish SageMaker wait time was that fast. Right now we have to use EC2 for any 'real' use cases with customers.)

No need to respond. But please improve zero-to-one start time across SageMaker services. Thanks.

### Other Details

_No response_

Contributor guide

Open the contributing guide

Research direction

Start by reviewing the Amazon SageMaker Studio launch flow described in the issue, including MLflow server creation, model registration, and endpoint creation. Measure the current zero-to-one times for these stages and identify the scope of the broader SageMaker services request. Done means materially shorter startup times, including for CPU-only endpoints.

Written by the indexing model from the issue text.

Assessment

Tech stack
aws
Domain
cloud, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.