Latency Percentile Calculator

How to use

1. 1Click a preset to load an example latency shape (web API, database query, gRPC, or ping), or paste your own samples into the input.
2. 2Pick a unit. Auto detects from the median of the input. Override with ns, µs, ms, or s if the auto pick is wrong.
3. 3Read p50, p75, p90, p95, p99, and p99.9, plus the min, max, mean, and standard deviation. The headline shows p95 in the unit that best fits the data.
4. 4Scan the histogram to see how the distribution is shaped. The bucket containing p99 is the one most worth investigating.
5. 5Optionally type an SLO budget and pick its unit. The tool reports the share of samples at or below the budget and explains the result in plain language.
6. 6Click Copy report to grab a plain-text summary with all percentiles, summary stats, and the SLO check, ready to paste into a runbook, dashboard caption, or pull request comment.

About this tool

Latency Percentile Calculator turns a list of raw latency measurements into the tail-latency numbers SREs and backend engineers actually publish. Paste a list of samples (separated by spaces, commas, tabs, or newlines), pick a unit or let the tool auto-detect from the median, and read p50, p75, p90, p95, p99, and p99.9 alongside the min, max, mean, and standard deviation. The percentile method is NIST type 7 linear interpolation, the same definition NumPy percentile, pandas .quantile, and SQL PERCENTILE_CONT use, so the numbers match what a Jupyter notebook would give you. A log-scaled histogram with twelve buckets shows the shape of the distribution so the long right tail does not collapse into a single bin, with a visual bar comparing the share in each bucket. An optional SLO budget input returns the share of samples that meet a target latency, with a one-line interpretation explaining whether the service has three-nines headroom, just passes p95, or is missing the budget entirely. Four built-in presets demonstrate four typical shapes: a fat-tailed web API in milliseconds, a tight OLTP database query distribution, an internal gRPC service in microseconds, and a ping round trip with a single packet-loss spike. The tool tolerates messy paste: it strips ms, s, µs, and ns suffixes, handles thousand separators, and silently ignores header rows or other non-numeric tokens (reporting how many were skipped). A copy-friendly plain-text report condenses the percentiles, summary stats, and SLO result into a runbook-ready block. Useful for SREs sizing alerts, backend engineers tracking p99 regressions, database administrators investigating slow queries, mobile and web developers reading dev-tools timing exports, growth teams sanity-checking dashboard math, and students learning what tail latency means in practice. All math runs locally in your browser; the latency samples you paste never leave your device.

Free to use. Works in your browser. No signup, no login.