Percentile (p95, p99)

A percentile tells you the value under which a given share of observations fall. If p95 response time is 800 ms, then 95% of requests finished in 800 ms or less, and the slowest 5% took longer. The median is p50.

Why it matters

Averages hide the pain of slow requests. If 98 requests take 100 ms and 2 take 10 seconds, the average is about 300 ms and looks fine, yet some users had a terrible experience. High percentiles (p95, p99) expose that tail. They matter more in services where one page view triggers many backend calls, because the chance of hitting at least one slow call grows.

Example

Ten response times in ms: 90, 95, 100, 105, 110, 115, 120, 130, 140, 2500.

average = 350 ms
median (p50) = 112.5 ms
p90 (nearest-rank method) = 140 ms

The median says typical users are fine, while the tail shows a serious outlier.

Common mistakes

• Averaging percentiles from several servers. They cannot be combined that way, so merge histograms instead.
• Using too few samples, so p99 swings wildly.
• Alerting only on the mean.

Related

• Latency
• SLI
• Metrics
• Apdex

यह पेज सामान्य जानकारी है, जिसे हमारी संपादकीय टीम ने लिखा है और दस्तावेज़ीकरण बदलने पर अपडेट किया जाता है। सर्वर और प्रदाता अलग-अलग होते हैं, इसलिए बदलाव पहले सुरक्षित तरीके से जाँचें। देखें: संपादकीय नीति और अस्वीकरण.