Announcing SOCI-indexed images for faster SageMaker AI jobs
Seekable OCI (SOCI) indexing for Amazon SageMaker AI training and processing jobs takes full container image download off the critical path. SageMaker AI lazy-loads your image to start containers in seconds instead of minutes, with no changes to your job definition. See the benchmarks and how to adopt it.
First seen by Oriel on .
AI summary
AWS adds Seekable OCI (SOCI) lazy loading to SageMaker AI Training and processing jobs, reducing container startup from minutes to seconds while leaving job definitions unchanged.
- SageMaker AI checks Amazon ECR for a SOCI index, fetches startup files first, and streams other image contents as they are accessed; failed lazy loading falls back to a standard download.
- Across images from 1.8–35.4 GB, lazy loading took 3–11 seconds versus up to 7 minutes 17 seconds for standard downloads, with reported improvements of 8x–46x.
- Training time per epoch matched after the first epoch; lazy loading added 3–16 seconds to that first epoch, while tested jobs finished 1–6 minutes sooner overall.
- Users can switch to pre-indexed AWS Deep Learning Containers with a -soci tag or index custom images using containerd and the SOCI snapshotter; indexing can also be automated with CFN ECR SOCI Index Builder.
- Lazy loading requires Amazon ECR private repositories, offers smaller gains for images below roughly 4 GB, and may provide little improvement when startup code reads most of the image.
#amazon sagemaker #seekable oci #container images #machine learning
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| AWS Machine Learning Blog | 4fef4e2052ceafe5be77506f74414eb6efbd5cd7 | https://aws.amazon.com/blogs/machine-learning/announcing-soci-indexed-images-for-faster-sagemaker-ai-jobs/ Read discussion |