Mind the Spike
Benchmarking for Worst-Case Execution Time in Realtime Code
What’s the longest your audio callback will ever take?
On modern CPUs, execution time is anything but consistent. Cache misses, pipeline stalls, thread contention, and OS jitter introduce variability—and occasionally, catastrophic outliers. These rare spikes are exactly what can break realtime audio, yet they’re nearly impossible to capture with conventional benchmarking. After all, how do you know the longest time you saw… really _was_ the longest?
This talk introduces Extreme Value Theory (EVT) as a practical framework for analyzing the worst-case execution time of realtime-critical code. We’ll look at how to apply EVT to benchmark data, how to interpret the results, and how this method can reveal serious system-level issues that may only surface once in a million callbacks.
Attendees will leave with actionable techniques for measuring, analyzing, and reasoning about the statistical edge cases of their code’s timing behavior—gaining stronger confidence in the robustness of their audio systems.
Christian Luther
Christian is the founder of Playfair Audio and a seasoned audio software developer from Hannover, Germany. With a background in electrical engineering and years of experience at companies like Kemper Amps and Sennheiser, he’s spent his career building tools that make audio production more intuitive. Through Playfair Audio, he focuses on solving the technical challenges so others can focus on being creative.