Achieving Excellence: SLO Thresholds That Transform Service...
summary
Netflix Open Connect serves 230 million members from over 18,000 OCAs and measures QoE with three concrete SLIs: play delay, network rebuffer counts, and normalized bitrate. Thresholds come from intuition, historical distributions (IQR math yields an example SLO of 87.5% normalized bitrate), surveys that show users quit after 2–3 rebufferings, and scalable in‑app feedback plus A/B experiments. Alerts detect regressions (example drop on Jan 10) and trigger ISP outreach tied into an ISP speed index (61 countries).
AI-written summary · about
chapters (9)
- 0:00SLOs and SLIs
- 1:38Netflix CDN Open Connect
- 5:11Quality of experience metrics
- 9:31Adaptive streaming trade-offs
- 9:51Setting SLO thresholds (level 1)
- 10:54Level two method using historical data
- 14:42Level three method using surveys and A/B tests
- 19:35Alerts with SLO thresholds
- 21:22Key takeaways
related talks
45mKeynote Address: Netflix: Streaming Entertainment to 200 Million Members Around the World
03mImproving ABR at Netflix
31mMastering Near-Real-Time Telemetry and Big Data: Invaluable Superpowers for Ordinary SREs
19mUsing application informed pacing to be a friendly internet neighbor
37mUsing SLOs for Continuous Performance Optimizations of Your K8s Workloads
22mProviding better video experiences for the next billion users.
33mThe Challenges of Testing Video Playback at Scale
14mARTEMIS: Adaptive Bitrate Ladder Optimization for Live Video Streaming