Powering 100+ million daily users
summary
Architecture that serves 100+ million daily users and targets a 250 ms goal for the card head uses layered caching, regional replication and predictable partitioning: Redis partitioning across multiple instances with consistent hashing and tolerant cache misses, prefetching guided by ML to cut wasted fills, and tunneled connections using SignalR then WebSocket fallback to avoid TLS latency. Built rest-over-WebSocket framing, scaled from 1M to tens of millions, hit hotspots from tenant IP affinity and adjusted the comms stack.
AI-written summary · about
chapters (15)
- 0:00Start
- 0:21Powering 100 Million Daily Users
- 1:41Core Mission of Microsoft
- 1:51What Is Productivity to You at Work
- 4:53Scalable Architecture
- 7:32Cache Resource
- 13:58Disaster Recovery
- 23:19Machine Learning Models
- 33:17Rest over Websockets
- 34:52Monoliths
- 35:02Micro Services
- 35:41Organized Monolith
- 37:32Mediator Pattern
- 38:53Designing for Performance
- 39:26Scaling Engineering
related talks
53mPowering 100+ million daily users
18mScaling APIs to Billions: Extreme-Scale Reliability
47mScaling: from 0 to 25 million users
26mHow We Prepped Our Infrastructure for a Super-Bowl Sized Traffic
51mScalable architecture from the ground up
38mHighly Scalable Consumer Applications using Cloud Native Open Source Tools
23mDesigning Robust and Scalable Distributed Applications
51mScaling Wix with microservices architecture