TwitterSpaces with Andrey Akinshin

This is a recording of the TwitterSpaces discussion with Andrey Akinshin on December 5th, 2021. We were talking about performance benchmarking, statistics, performance regression analysis, and more. Show notes: https://gist.github.com/chadbrewbaker... Highlights: Benchmarking very much depends on your business goals. There is no universal benchmarking method/approach that can be applied everywhere. Visualizing distribution plots is very useful since performance results are rarely a normal distribution, sometimes multimodal distributions. Regression analysis can be parametrized using four main parameters: false-positive rate, false-negative rate, effect size, number of measurements. Try to eliminate big sources of noise in the benchmarking environment, e.g. background processes, dynamic frequency scaling. If you do not control the environment, use statistical methods.