Announcing SOCI-indexed images for faster SageMaker AI jobs
According to AWS. Amazon Web Services announced Seekable OCI (SOCI) indexing for Amazon SageMaker AI Training and processing jobs to lazy-load container images.
- • SOCI indexing takes full container image download off the critical path of every Amazon SageMaker AI job.
- • SageMaker AI lazy-loads the container image by fetching only the files needed to start and streaming the rest in the background.
- • Benchmarks showed download times dropping from several minutes to 3-11 seconds across multiple image sizes.
Overview of SOCI Indexing
Amazon Web Services announced Seekable OCI (SOCI) indexing for Amazon SageMaker AI Training and processing jobs. The technology takes full container image download off the critical path by using a SOCI index to lazy-load container images, allowing jobs to begin in seconds instead of minutes.
Measured Performance and Adoption
Benchmarks across compressed image sizes ranging from 1.8 GB to 35.4 GB showed that download times dropped significantly while maintaining training performance after a minor first-epoch cost. Users can adopt the feature by pointing to a pre-indexed AWS Deep Learning Container or adding an index to their own image without modifying job definitions.
- 1. aws.amazon.comOpen source