API Response Time Analyzer

Percentile latencies describe the distribution of response times. p50 (median) means 50% of requests complete in that time or less. p90 means 90% of requests are that fast or faster — 10% are slower. p99 is the '99th percentile': 99% of requests complete in that time, only 1% are slower. p99 is critical for user experience because it captures your worst-case 'tail latency'.

The mean is easily skewed by outliers. If 99 requests take 10ms and 1 request takes 10,000ms, the mean is ~110ms — but the p99 is 10,000ms. Users experience individual requests, not averages. Always monitor p95 or p99 alongside the mean for a realistic picture of performance.

General guidelines: under 100ms feels instant, 100-300ms is acceptable for most API calls, 300ms-1s starts to feel slow, over 1s significantly impacts user experience. For background jobs or data export endpoints, higher latency is acceptable. Real-time features (chat, live search) should target under 50ms.

Common causes: garbage collection pauses, database lock contention, cold cache misses, N+1 query problems, connection pool exhaustion, large payload serialization, external API slowness. Profile at the p99 level — optimizing the median won't fix tail latency if the root cause is intermittent.

Provide a JSON array where each object has at minimum a `url` string and `duration` number (milliseconds). Optionally include `size` (bytes) and `status` (HTTP status code). Example: [{"url": "/api/users", "duration": 45, "size": 1200, "status": 200}]. You can export this format from many APM tools.