Measuring Latency for Monitoring and Benchmarking purposes is notoriously difficult. There are a lot of pitfalls with collecting, aggregating and analyzing latency data.
In the talk, we will make an effort to visit this topic from a top-down perspective and compile known complications and best-practice approaches on how to avoid them. This will include:
- Measurement Overhead
- Queuing effects – Coordinated omission
- Histograms for Aggregation and Visualization
- Percentile aggregation
- Latency bands and burn-down charts
- Latency comparison methods (QQ Plots, KS-Distance)