Percentiles and coordinated omission
On This Lesson 4 topics
This lesson is planned. The outline below lists the topics it will cover. The full lesson has not been published yet.
In this lesson
report latency honestly.
Planned topics
- Averages hide the slow tail; report p50, p95, p99 and the max, plus throughput and error rate.
- Percentiles cannot be averaged across machines; combine histograms.
- Coordinated omission: a closed-loop tool that waits for a slow response stops sending, so it records fewer slow samples and the result looks better than reality.
- Use arrival-rate (open) executors or tools that correct for it; state the method in the report.
Draft outline details
Demonstration
a stall of 10 seconds: a closed loop records one slow request, an open model records every request that should have been sent.
Check your understanding
Why is average latency a poor report?
Show answer
It hides the slow requests users actually feel.
Practice
Apply this lesson with the module's lab, exercises and quiz.
Marking a lesson complete is optional and you can undo it any time. Your progress stays in this browser.