Skip to main content

Calculations run on this device. Scenario values are not sent anywhere — the engine is fully local.

inference calculator · free

Back-of-the-Envelope Estimator

Turn QPS, fan-out, and per-request cost into a defensible capacity number.

How many servers does this traffic estimate really need?

Engine 1.0.0 · calculation
2,000 QPS
3 leaf calls

One incoming request multiplies into this many leaf requests.

4 ms
50 %

Utilization ceiling you refuse to plan past.

3machines (16c) at 50% headroom
Leaf traffic
6,000 QPS
Cores busy at peak
24
Cores to provision
48
Rule used
Little's LawL = λW

Concurrency = rate × latency. Provision for the leaf tier, not the edge: fan-out is the multiplier people forget.

Method

  • Little's Law: concurrent leaf work = leaf QPS × per-call core seconds.
  • Provisioned cores divide busy cores by target headroom; machines assume 16 cores each.
  • Fan-out is applied before capacity math — the edge QPS is never the provisioning input.