OCIv2: Container Images Considered Harmful
summary
Argues that tar‑based OCI layers cause waste and duplication (touching a 10 GB file can double an image, deletions add 512‑byte whiteout headers), and proposes OCIv2 images: a content‑addressable inode manifest pointing to chunked blobs so identical files deduplicate across images. Uses hamachi/emoji2 to demo an OCI store, shows ref links for copy‑on‑write rootfses and near‑instant extraction when only one file changes. Implementation works but transfer compression still favors monolithic tar/gzip, so transfer‑side fixes remain.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:40Ideal Image Format
- 4:37Tar
- 9:40Transfer
- 14:10Alternative
- 23:05Demo
- 36:10C8R
- 39:00Where to get it
- 39:45Next steps
related talks
10mLightning Talk: OCI Images for More Than Containers, Jon Johnson, Google
25mStreamlined Efficiency: Unshackling Kubernetes Image Volumes for Rapid AI Model...
25mStop Re-Downloading Your Container Images: Content-Based Layers with Chunkah
36mModifying the Immutable: Attaching Artifacts to OCI Images
35mStore AI/ML models efficiently with OCI Artifacts
36mIt’s Complicated: Relationships Between Objects In OCI Registries
40mNode-Level Optimization in Large Cluster Powered by Containerd Plugin
30mPod fails to start? Consider OCI image compatibility