IP Clusters - Scaling Zopim's Frontend Node Servers
summary
Claim: Zopim scales a global realtime Node service handling 500,000 concurrent connections and 3 billion monthly impressions by avoiding hardware balancers and using an open source IP cluster that manipulates least-significant-bit netmask routing and iptables to implement sticky routing. Implementation details include using an SSL terminator (start), detecting worker overloads and memory with 2bz, graceful worker retirement, hot deploys, automated restarts and heap diffs; demo flops are mentioned but the system is deployed and solved practical scaling pain.
AI-written summary · about
chapters (6)
- 0:00Intro
- 0:25Outline
- 4:01cluster module
- 12:15Remaining Issues
- 16:14Retiring Workers
- 22:45IP Cluster - Future
related talks
22mMulti-tiered Node.js architectures
38mCloud Scale Erlang
47mReal-world scaling with Scala & Akka
37mBuilding Fast Scalable Game Server in Node.JS (updated)
45mKeeping it Real(time)
43mRealtime performance at scale (...)
31mProgramming a Node Cluster Like It's One Really Big Computer
18mHigh-Load Systems: Challenges in Social Network Development