Availability, Latency and Cost: Withstanding Regional Outages
summary
Netflix describes an active‑active multi‑region deployment with DNS/SSL terminating proxies and regionally isolated change rollout, backed by algebraic availability models and incident playbooks. Failover time drops from ~45 minutes to failover under 10 minutes using "nimble" tooling; cost models show sublinear throughput overhead and tradeoffs between RF=all vs RF=2 (e.g., three regions ≈ 50% throughput + 3× DB cost, RF=2 designs can hit ≲50% overhead with more operational complexity). Caches, async replication and steering heuristics complete the design.
AI-written summary · about
chapters (4)
- 0:00Start
- 7:15In general, freedom and rapid recovery is better than trying to prevent error. We are in a creative business, not a safety-critical business.
- 10:41Distribution of Change Number of Regions Balance of Traffic
- 32:20Closing Thoughts
related talks
44mAvailability, Latency and Cost: Withstanding Outages
44mTechniques Netflix Uses to Weather Significant Demand Shifts
52mMy Children Will Never Deploy Active-Passive
21mHandling Traffic Spikes Through Automated Prioritized Load Shedding
20mBetter Resilience Through Self Tuning
18mLeveraging Multiple Regions to Improve Site Reliability:
27mBehind the Scenes: Running Critical Services @ Netflix
38mSecuring Distributed Cache: Achieving Secure-by-Default with Key Challenges &...