50: Service Level Agreements/Indicators/Objectives - Percentiles

In software reliability and availability, there are usually vague descriptions. When we say we offer good performance, what does it mean? This is why we use percentiles. Let’s think about response times, often used in SLOs, along with percentiles. In SLAs, we define X ms response time in the 95th percentile. To calculate this, put all response times and sort them from fastest to slowest. Pick the 95th one if you have 100 response times. (950th in 1000, and so on) The response time at that exact point defines your performance in the 95th percentile. For example, you can guarantee to serve in 200ms in the 95th percentile. The same applies to 99th and 99.9th percentiles. Pick accordingly from the sorted list. We also talk about the mean in response times. The mean response time is actually the 50th percentile. To find it, pick the middle one in the sorted list, and it shows the 50th percentile. For example, 200ms in the 95th percentile translates into 95% of all requests served faster than 200 milliseconds.

• Related Note(s):
  • 49: Accidental Complexity;
  • 48: System Design Index;
  • 47: Software Architecture Styles Index;
  • 39: Increasing Quality of Systems;
  • 36: Open Feedback Culture;
  • 24b1: Begin with The End in Mind;
  • 4b: Monitoring & Inspection to Learn;