Web application performance matters because users quickly abandon unresponsive sites out of frustration, but it is often difficult for developers to diagnose and remedy performance problems. Due to their pervasive event-based concurrency, existing web profiling tools rarely provide actionable output that can guide web developers to optimization opportunities.

In this talk, I describe a new technique for causally profiling event-driven programs that I implement in BCoz, a prototype causal profiler for web applications. BCoz identifies and quantifies optimization opportunities in web applications, letting developers concentrate on optimizations that matter on the metrics they care about. BCoz is primarily implemented in browser-agnostic JavaScript, and only requires a few changes to Google Chrome to leverage its existing sampling-based profiler and timestamped event loop