Response Time Percentile Calculator

Work out response time percentile instantly with clear inputs, formula shown and shareable results.

About Response Time Percentile Calculator

The ratio between p99 and p50 characterises the tail. A healthy web service sits around 3 to 5 times; ratios above ten usually point at queueing, garbage collection pauses, cold caches or one slow dependency. Because a single page view may issue dozens of requests, the p99 of individual calls becomes the typical experience of a page, which is why tails matter more than medians.

Formula

Tail characterisation

tail ratio = p99 / p50; requests above SLO per second ~ rate x 1% x fraction of the p99 tail beyond the SLO

Worked example

With the default values below, Response Time Percentile Calculator returns p99 to p50 ratio of 7.53. Change any input above and every figure here is recalculated in your browser.

Inputs used

Median (p50) latencyminimum 0.01 ms: 85 ms
p99 latencyminimum 0.01 ms: 640 ms
Request rateminimum 0.01 req/s: 1,800 req/s
Latency SLO thresholdminimum 0.01 ms: 500 ms

Results

p99 to p50 ratio: 7.53 x
Requests likely exceeding the SLO per hour: 14,175
p99 headroom against the SLO: -140.0 ms
Distribution assessment: Typical web-service tail

Step by step

1Values used
2Tail characterisation
3p99 to p50 ratio= 7.53 x
4Requests likely exceeding the SLO per hour= 14,175
5p99 headroom against the SLO= -140.0 ms
6Distribution assessment= Typical web-service tail

Assumptions and limits

The result is derived only from the 4 inputs listed above. Anything not among them is not part of the calculation.
In particular, this calculator does not take inflation, tax, fees or charges or compounding frequency into account.
Figures are rounded for display to 2 decimal places, so totals built from several rounded parts can differ slightly from the unrounded arithmetic.

How these calculators are built and tested is described on the Methodology page. If a figure looks wrong, please report it.

Frequently Asked Questions

Why does the tail dominate user experience?

Fan-out. If a page makes 50 backend calls, the chance that at least one hits the p99 is about 40 percent, so most page loads contain a slow call.

Can percentiles be averaged across servers?

No. Averaging percentiles is mathematically invalid. Aggregate the underlying histograms or use a sketch such as t-digest or HDR histogram.