Dragonfly: Make Image Distribution Efficiently and Safely in Cloud Native
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Wednesday, 2020-11-18· at 17:45· Intrado Virtual Event Platform,·
summary
Dragonfly is a P2P image distribution system built from dfdaemon/dfget/supernode that schedules downloads at block granularity (default 4 blocks per schedule) to offload origin servers and keep download time flat as concurrency rises; community metrics show 5200+ stars, ~70 contributors, 50+ companies using it. A new subproject introduces a metadata/data split: images expose a tiny metadata layer and fetch chunked file data on demand (chunks deduplicated, LZ4 supported), yielding container starts under ~4s and metadata pulls under 1s.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:21Agenda
- 0:50Introduction
- 8:37How Dragonfly Participates in Image Distribution
- 15:23Three Consequences of the Defect
- 19:59Architecture
- 26:52Contact Information
related talks
37mIntro: Dragonfly
34mDeep Dive: Dragonfly
29mAccelerates Image Distribution In Multi-Cluster With Dragonfly
31mThe New Evolution of Image Distribution System Dragonfly
36mDragonfly: Intro, Updates and AI Model Distribution in the Practice of...
30mDragonfly V2.5.0 - Intro, Updates, Data Distribution in AI Infrastructure
29mPull Once, Scale Everywhere: Fixing Image Pull Bottlenecks With Dragonfly
34mThree Approaches to Speed up Image Distribution in Cloud Native Era