percentile latency
also: p50, p95, p99, p99.9, tail latency
Percentile latency measures the maximum response time experienced by a certain percentage of requests in a system. For example, 95th percentile latency means 95% of requests completed within that time, while 5% took longer.
Percentile latency is a statistical metric used to understand the distribution of response times or delays across many requests. Unlike average latency, which can be skewed by outliers, percentile latency reveals what real users actually experience at different parts of the distribution.
Common percentiles include 50th percentile (median), 95th percentile, 99th percentile, and 99.9th percentile. For instance, if a web service has a 99th percentile latency of 500ms, it means 99% of requests complete in 500ms or less—but 1% experience delays exceeding that. The higher the percentile, the worse the outlier cases become.
In Linux monitoring and performance tuning, percentile latency is measured using tools like perf, LatencyTOP, or application-level monitoring. For example:
p50_latency=45ms p95_latency=120ms p99_latency=850ms