The Evolution of Twitter's Edge

on YouTube ↗28m

" summary

Twitter runs Envoy at the edge for 100% of requests, handling hundreds of billions per day, replacing a bespoke proxy (TSA) with roughly 2x throughput and better latency. An internal Scala control plane serves xDS, custom C++ filters add Twitter metrics and a global connection hash, and contributions include per-connection rate limiting (cuts TFE tail latency ~20%) and TLS fingerprinting for bot detection; TLS 1.3 adoption drops handshake time ~38% and home-timeline refresh latency ≈30%.

AI-written summary · about

" chapters (9)

1. 0:00Introduction
2. 1:54Availability and Reliability
3. 4:40Pre2014
4. 9:22Envoy
5. 12:46T3 Proxy
6. 14:51Control Plane
7. 18:08TLS 13 Performance
8. 19:01Bot Mitigation
9. 23:14Whats Next

" related talks

• 42mScaling to Billions of Real-time Recommendations DailyAjeet Grewal · Scale By the Bay 2017
• 41mInside Twitter's Realtime Delivery of Tweets for Enterprise CustomersLisa White · Strange Loop 2017
• 22mThe Open Source Technology Behind a TweetChris Aniszczyk · LinuxCon North America 2012
• 22mRebuilding Twitter’s public APISteve Cosenza · Scale By the Bay 2020
• 27mFixing Twitter: Improving the PerJohn Adams · Velocity 2009
• 26mRunning Envoy at the EdgeDerek Argueta · EnvoyCon 2018
• 42mPowering 100+ million daily usersRezaul Hoque · NDC Oslo 2019
• 06mInterviewJulien Delange · Scale By The Bay 2018