$linuxjunkies
>

p99

also: 99th percentile, p99 latency, tail latency

The 99th percentile latency—the response time below which 99% of requests complete, used to measure system performance and tail latency.

P99 is a statistical measure that captures the latency threshold where 99% of requests fall below. If a service has a p99 latency of 100ms, it means 99 out of 100 requests complete in 100ms or less, while up to 1% of requests may be slower.

P99 is particularly useful for identifying performance problems that affect a small percentage of users. While average latency might look healthy, high p99 values reveal slow outliers—such as cache misses, disk I/O spikes, or garbage collection pauses—that degrade user experience.

Related percentile metrics include p50 (median), p95, and p99.9. For example, a web service might report: p50=20ms, p95=50ms, p99=200ms, p99.9=500ms, showing that most requests are fast but a small tail is significantly slower.

Related terms