Advertisement
Averages lie. p99 latency is what users actually experience at the tail.
What you're seeing
Average latency hides the tail. A service with avg 100ms and p99 of 2000ms means 1% of users wait 20× as long as average. Often the 1% is your biggest, most-engaged users.
Track p50/p95/p99/p999. SLOs almost always set on tail percentiles, not averages.
★ KEY TAKEAWAY
Average lies. p99 latency = what 1% of users experience. SLOs use percentiles, not averages.
▶ WHAT TO TRY
- Switch between distributions — bimodal and heavy-tail have very different p99s with similar averages.
- Click Resample to see new sets.