Using the Internet as Your Load-Balancer
summary
Compares DNS-based steering versus BGP anycast and gives operational recipes: use EDNS Client Subnet is patchy so resolver IPs mislead, prefer deterministic BGP tie breakers like local-pref and AS path, and manipulate reachability with BGP communities to stop export or limit region. Describes incident moves: announce a /20 to override a /24 then withdraw the /24, measure with RIPE Atlas probes and RUM, and implement BPF-based direct server return to shed sessions while preserving most traffic.
AI-written summary · about
chapters (29)
- 0:00Introduction
- 1:58DNS or Routing
- 2:28DNS
- 2:54IP Allocation
- 3:41IP Location
- 7:09Concerns
- 8:40Recap
- 9:03Unicast
- 9:31Anycast
- 10:34TCP Handshake
- 11:14How to get stable connections
- 11:54ECMP
- 12:13Consistent hashing
- 12:40Routing algorithms
- 13:09BGP
- 18:14Communities
- 18:58Summary
- 19:17SLOS
- 20:09Routing Equipment
- 20:43Flow Analysis
- 21:54Probes
- 22:20RIPE
- 22:57Real User Monitoring
- 23:52Debug Output
- 24:02Too Much Load
- 24:52Anycast Backup
- 26:30Load shedding
- 26:44Direct server return
- 28:22Wrap up
related talks
30mKeeping the Balance: Internet-Scale Loadbalancing Demystified
46mKeeping the Balance: Load Balancing Demystified
41mImpact of Network Automation
11mRunning BGP in Data Centers at Scale
34mHyperRouter: Lessons Learnt from Building an L4 Load...
35mIP Anycast: DC/OS Networking for the Modern Datacenter
15mCost-effective and Reliable Global Internet Peering with Programmable Switches
13mHow the BGP Protocol Makes the Internet Work — and Could Make It Disappear