p50 · p95 · p99
A visual guide to performance percentiles and why they matter more than averages.
“Users don’t experience your average. They experience your worst moments.”
Latency Distribution
Most requests are fast, but tail latency reveals how your slowest users experience your service.
Common Percentiles
| Percentile | Meaning |
|---|---|
| p50 | Half of requests are faster than this |
| p90 | 90% of requests are faster |
| p95 | Only 5% of requests are slower |
| p99 | Only 1% of requests are slower |
| p99.9 | 1 in 1,000 requests is this slow or worse |
Percentile Formula
Where p is the desired percentile and N is the total number of sorted observations. The result is the index into the sorted array. Interpolate between adjacent values for non-integer indices.
100 Users
Each square = 1 request. Scale up to see how tail latency affects more users.
Why Averages Lie
100 API requests with a realistic distribution. The average blends them into a single misleading number. Percentiles show what each tier of user actually experiences.
Drag the outlier slider, then scale up the traffic to see how many real users are affected.
Why Percentiles Matter
Same data, different story. Here is what you would miss by only looking at averages.