How We Achieved Even Load Balancing on the Frontend Cluster
summary
Describes achieving very even load balancing across a frontend cluster of 200+ servers by implementing automatic weight tuning for the load balancer, reducing CPU usage spread to about 2.5% at traffic peak. Names concrete results: saved roughly 40–50 servers and improved mobile site response times; the implementation of the described algorithm will be released as open source.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 3:14Load balancing needed
- 7:52Load balancing algorithms
- 10:00Manual tuning challenges
- 12:52Automated balancing
- 21:51Results and plans
- 22:56Q&A
related talks
40mGood, Bad, and Ugly - the Art of Load Balancing at Scale
46mKeeping the Balance: Load Balancing Demystified
11mLightning Talk: The Adaptive Edge: Balancing Cost Efficiency, Usability, and Performance Tradeoffs
24mLoad-Balancing Node.js in Production
18mDeveloping a Custom Load Balancer to serve traffic at scale
37mBuilding the Multi-Cluster L7 Load Balancer
47mA Frontend Server, Front to Back
24mReliability Is Not Just About Balancing Loads