Intro: Dragonfly
summary
An open-source P2P image and file distribution system built for production scale, Dragonfly uses a P2P supernode + DFget clients architecture, block-level transfers, dynamic compression and an in-memory file system to cut cross-region bandwidth and registry load. It runs at 1.5 billion downloads/month inside Alibaba, keeps average pull cost flat as concurrency grows (native registries spike), and shows ~3x faster deployments in adopters. Refactored in Go, it supports HTTP registries, CDN seeding, encryption and plugin-based scaling.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 3:32Milestones
- 7:33Status of Dragonfly
- 9:29Dragonfly Community
- 13:25Supernode Architecture
- 14:45Dispersity Theory
- 19:25Compression
- 21:25Storage Manager
- 25:25Demo
- 27:25Data comparison
- 29:25Scalability
- 31:25Repository
- 33:25Supernode
- 35:25Syncing
related talks
25mDragonfly's CNCF Graduation Journey and Large-Scale AI Model Distribution in Pra...
30mData @Scale Seattle Samir Goel
30m🛠️ Filecoin data onboarding with Delta
30mPeernet - an open p2p network standard / implementation
14mFalconFS: Distributed File System for Large-Scale Deep Learning Pipeline
30m🛠️ Filecoin data onboarding with Delta
15mThe State of Filecoin
45mA Distributed File System for Secure P2P Applications
