SYSTEM Cited by 1 source
Amazon SOCI (Seekable OCI)¶
SOCI (Seekable Open Container Initiative) is an open-source index format, maintained as an AWS-originated CNCF sandbox project, that lets a container runtime lazily fetch only the filesystem layers it actually needs at startup rather than pulling the full multi-gigabyte image up front. An SOCI index — computed once, stored in ECR — describes the internal layout of the image so the runtime can start the container immediately and stream filesystem data on demand.
Why it exists¶
Container images in practice drift into the multi-gigabyte range as teams add dependencies (especially deep-learning runtimes: CUDA + PyTorch + model weights). On serverless compute substrates that provision instances on demand (e.g. SageMaker Jobs, Lambda), pulling the full image is a cold-start tax linear in image size. SOCI converts that O(size) tax into O(working-set-at-startup). See concepts/lazy-container-image-loading.
In production — Lyft / LyftLearn 2.0¶
Lyft's LyftLearn ML platform adopted SOCI for JupyterLab notebooks on SageMaker Studio and cut notebook startup times by 40–50% (Source: sources/2025-11-18-lyft-lyftlearn-evolution-rethinking-ml-platform-architecture). At migration time, SOCI "wasn't available" for SageMaker training / batch jobs — Lyft used SageMaker warm pools to cover that gap for the most latency-sensitive training jobs (models retraining every 15 minutes).
Seen in¶
- sources/2025-11-18-lyft-lyftlearn-evolution-rethinking-ml-platform-architecture — 40–50% JupyterLab notebook startup reduction.