Streamlined Efficiency: Unshackling Kubernetes Image Volumes for Rapid A...
summary
Claims a way to mount OCI image volumes without repackaging or full pulls by streaming remote objects via a reference list (source path, mount path, ETag, size) stored as an OCI layer annotation. Implementation maps file IO to remote range GETs and uses an overlayBD block-device backend for sector-level on‑demand reads. Packaging a 22 GB dataset of ~700k files drops from ~4 hours to under 2 minutes to build the mount, and throughput for many small files outperforms traditional OCI image mounts and goofys. mount without pulling ETag validation
AI-written summary · about
chapters (9)
- 0:00Kubernetes efficiency intro
- 1:04Image startup challenges
- 3:34OCI registries for data
- 6:48Registry volume limits
- 8:59Large dataset packaging costs
- 11:06Introducing ELink
- 16:31ELink mount points
- 20:32ELink vs OCI performance
- 22:45Conclusion & Q&A
related talks
37mCache Me If You Can: P2P Image Sharing in Kubernetes with Spegel
26mPains and Gains: Lessons From Orchestrating Volumes for Remote Storage in Cloud Native AI Scenarios
34mFast Image Pulls Using IPFS And Opportunistic Caching
40mNode-Level Optimization in Large Cluster Powered by Containerd Plugin
30mSpeeding Up Analysis Pipelines with Remote Container Images
29mReliable and Cost-Effective: Open Storage Strategies for Kubernetes
29mSo, What If I Don’t Want My Persistent Storage To Be Yet Another Bindmount?
25mContainer Attached Storage: Open Source Extensions: Kubernetes + Data