Percentile (p50, p95, p99)
The value below which a given share of measurements fall. The p50 (median) latency is what a typical request sees; the p99 is the latency that 99% of requests beat, so 1 request in 100 is slower. Averages hide slow outliers, so SLOs and alerts usually use high percentiles.
Observability
In Quorum#
Before Percentiles research the HUD shows mean latency; after it, the HUD shows the p99 and each request kind shows its p50, p95, and p99. Event-loop stalls and GC pauses hit about 2% of requests, so they raise the p99 while the mean and the p95 barely move. With Alerting, the latency rule pages you when a kind's p99 stays above 4 times its Apdex target for 30 s. The Latency chart on the Dashboard draws the p99 next to the mean, so you can see how much the mean hides.