Keeping the Balance: Internet-Scale Loadbalancing Demystified
summary
Starts from a single A record and shows why DNS multi‑IP and short TTLs still fail for health and control, then moves to network load balancers using 5-tuple hashing (and Google's Maglev consistent hashing claim to avoid connection resets). Explains proxy vs DSR direct server return (DSR cuts egress LB bandwidth), BGP/anycast and geo‑DNS limits, layer‑7 reverse proxies, and algorithms like choice‑of‑two to avoid thundering‑herd effects. Practical tradeoffs, costs, and where sidecar/client load balancing fits.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 6:37Load balancers
- 10:33Network devices
- 13:00Capacity
- 17:30Multiregion
- 21:47Application Load Balancing
- 25:48Load Balancing Algorithms
related talks
46mKeeping the Balance: Load Balancing Demystified
29mLoad balancing: an in-depth study to scale @ 80K TPS
29mUsing the Internet as Your Load-Balancer
53mHTTP Traffic Load Balancing
33mCloud Native Load Balancing From Scratch
40mGood, Bad, and Ugly - the Art of Load Balancing at Scale
52mGlobal web resource load balancing across geo‑distributed infrastructure
33mCloud Native Load Balancing From Scratch