Scaling the Public Edge: Approaches to Application Load Balancing

Speaker: Chris Woodfield, Twitter
Nearly every public website and application today scales beyond a single web server, and the techniques for scaling websites and other applications on the public Internet vary far and wide, and often change as sites approach "Internet scale." In this panel discussion, we will discuss various operators' and vendors' approaches and challenges to load balancing large-scale web sites/apps, both at the network and service level, and how increasing growth and new technologies have guided their scaling models and strategies. Sample panel questions: How has your site's load balancing needs changed at various points of scale? What systems or solutions have proven to navigate these changes with adjustments, and what approaches required a complete overhaul? What systems or solutions have proven to be the most scalable and manageable? What features are you seeing as missing from current solutions in the market today? What pros and cons of appliance-based vs. virtualized vs open-source load balancing solutions have you encountered? What automation challenges have you encountered and what solutions have you adopted? Technologies in the discussion scope will include: Layer 7 (proxy-based) vs. Layer 4 (NAT-based) load balancing, "inline" vs. DSR High Availability methods: Active/Standby, Active/Active, N+1 clustering Load sharing and failover with multiple VIPs: DNS round-robin, IP route injection, ECMP/anycast approaches Commercial versus open source solutions Appliance vs software vs. virtualized solutions