Performance engineering from kernel analysis to AI: Adrian Cockcroft’s take
Over its five-year history, P99 CONF has hosted quite a few speakers who’ve offered pointed takedowns of the namesake metric. The post Performance engineering from kernel analysis to AI: Adrian Cockcroft’s take appeared first on The New Stack .
Adrian Cockcroft, a renowned performance specialist, spoke at P99 CONF, a conference dedicated to performance engineering. Cockcroft, with his extensive experience at Sun Microsystems, Netflix, eBay, and Amazon, emphasized the importance of digging deeply into performance problems. He recalled how, at Sun, engineers would often rely on system metrics like vmstat, but lacked a clear understanding of what these numbers meant.
This led him to read the kernel source code and write performance books, "Sun Performance and Tuning" and "Resource Management".
Today, Cockcroft uses modern tools like AI to analyze performance issues. He has developed tools to better understand response time distributions, rather than relying solely on percentiles. Percentiles, like P99, can be misleading when analyzing the performance of modern web services. This is because they do not account for the distribution of response time peaks. Cockcroft believes that a single number cannot accurately represent the complexity of real-world performance issues.
To address this, Cockcroft has developed a custom tool that identifies peaks in response time distributions and tracks their fluctuations over time. This tool, implemented in R, is open-source and allows for a more nuanced understanding of performance. Cockcroft advises teams working on high-performance systems to start with a macro view to identify interesting patterns, then delve deeper to inspect individual slow requests end-to-end.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
