RPC Benchmark Percentiles Explained: p50, p95 and p99

Why averages hide the latency spikes that break trading bots, wallets and production blockchain applications.

Median

p50 describes the typical request but says little about congestion and provider degradation.

Tail latency

p95 and p99 reveal slow requests that trigger stale quotes, expired blockhashes and retry storms.

Sample design

Use enough samples across multiple time windows. Separate warm and cold connections and avoid coordinated benchmark bursts.

Decision making

Set application service-level objectives from tolerated tail latency and error rate, not a marketing average.

TierZero provides infrastructure engineering and 24/7 bot maintenance.