Resiliency Practices in Managing CDN (Content Delivery Network)
summary
Netflix Open Connect runs ~18,000+ OCAs in ~6,000 locations across 175 countries with ~70 IX sites on a multi‑terabit backbone, and validates resiliency with concrete exercises: stack failure (shut a line‑card at peak, observe t+~6m control‑plane convergence and client failover), stack overload (route 2x traffic to one stack for weeks to test OCAs and clients), and regional anycast/DNS withdrawal. Key metrics are average bitrate, play delay and rebuffering; Live (March 4 Chris Rock) is deployed on OCs and meets those resiliency checks.
AI-written summary · about
chapters (12)
- 0:00Netflix content delivery
- 2:23Open Connect appliances
- 5:16Global CDN footprint
- 7:20Streaming workflow
- 9:22Resiliency strategies
- 13:22Stack failure testing
- 17:09Client-side adaptation
- 21:28Stack overload testing
- 24:45Resiliency recap
- 26:42Anycast networking
- 29:27Live event infrastructure
- 31:00Philosophy and culture
related talks
43mScaling the Netflix Global CDN, lessons learned from Terabit Zero
43mBe Like Netflix - Building a CDN from Scratch
32mCDN Scaling Challenges
31mLive Q+A 1
11mIs the Internet Reliable Yet?
04mLoad balancing through the years
36mNetwork Machinery: A United-Front for Network Troubleshooting with CRDs
48mRethinking Connectivity