It's common knowledge that Twitter has been migrating from Ruby to JVM-based languages, but did you know we're also re-architecting the entire service? As Twitter has grown, we've found that a monolithic stack makes it hard to develop new features and maintain existing ones, so we've begun slicing our system into well-encapsulated, discrete services. I'll talk about how and why we're doing this (it's not because "Rails doesn't scale"—it does scale!), and the tradeoffs we've made as we optimize for performance and productivity. You'll walk away with new ideas for your own architecture and the know-how to scale from hundreds to millions while keeping uptime up and latency low.