Histogram for Response Time Latency — API & Database Tail Distribution
An API latency histogram buckets request durations into contiguous millisecond intervals to reveal right-skewed tail behavior, cache-hit vs. cold-read bimodal peaks, and SLA compliance.
Tip: Paste Excel or Google Sheets rows (Ctrl+V / ⌘V) directly into any cell.
| # | Label | ||
|---|---|---|---|
| 1 | |||
| 2 | |||
| 3 | |||
| 4 | |||
| 5 | |||
| 6 | |||
| 7 |
SLA Compliance % (<= T_sla) = (sum_{bucket <= T_sla} Req_i / Total Requests) * 100%. Tail error budget burn = 100% - SLA Compliance %.
Chart Design & Use-Case Tips
- 01SLA Compliance % (<= T_sla) = (sum_{bucket <= T_sla} Req_i / Total Requests) * 100%. Tail error budget burn = 100% - SLA Compliance %.
- 02API latency distributions are almost never normal bell curves; they are right-skewed log-normal distributions with a long tail caused by GC pauses, cold starts, or unindexed queries.
- 03A second bump in the 100–200 ms bin alongside a 15 ms peak signals a split between Redis cache hits and database fallbacks.
- 04Pair with the Box Plot for Server Latency Percentiles to track P50, P25, and P75 across microservices.
Frequently Asked Questions
Why is average response time misleading for web APIs?+
Because latency distributions are heavily right-skewed, a mean of 35 ms can hide a 500+ ms tail affecting thousands of checkout requests. Histograms and percentiles expose the full tail.
Can I paste Prometheus `histogram_bucket` deltas here?+
Yes, paste non-cumulative bucket counts from Prometheus, OpenTelemetry, Datadog, or CloudWatch Logs Insights directly into the workbench.
How do I share this chart in an architecture review?+
Click 'Export Vector SVG' or 'Export 300-DPI PNG' for a clean dark- or light-mode graphic.