Shadow Deployment Traffic Calculator

Work out shadow deployment traffic instantly with clear inputs, formula shown and shareable results.

About Shadow Deployment Traffic Calculator

Shadow deployment sends a copy of live traffic to a candidate model without returning its answers to users, so you get production-distribution behaviour with zero blast radius. The cost is real capacity: mirroring 20 percent of traffic means provisioning 20 percent more inference. Total mirrored volume determines whether the run collects enough events to compare distributions with confidence.

Formula

Shadow traffic

shadow rate = production rate x mirror share; total requests = shadow rate x duration in seconds; cost = requests x price per million / 1e6

Worked example

With the default values below, Shadow Deployment Traffic Calculator returns Shadow request rate of 80.00. Change any input above and every figure here is recalculated in your browser.

Inputs used

Production request rateminimum 0.01 req/s: 400 req/s
Traffic mirrored to the shadow modelaccepts 0.1–100 %: 20 %
Shadow run durationminimum 0.1 days: 14 days
Shadow inference cost per million requestsminimum 0: 12

Results

Shadow request rate: 80.00 req/s
Requests mirrored over the run: 96,768,000
Extra serving capacity required: 20.00%
Shadow inference cost: $1,161.22

Step by step

1Values used
2Shadow traffic
3Shadow request rate= 80.00 req/s
4Requests mirrored over the run= 96,768,000
5Extra serving capacity required= 20.00
6Shadow inference cost= 1,161.22

Assumptions and limits

The result is derived only from the 4 inputs listed above. Anything not among them is not part of the calculation.
In particular, this calculator does not take inflation, tax or compounding frequency into account.
Figures are rounded for display to 2 decimal places, so totals built from several rounded parts can differ slightly from the unrounded arithmetic.

How these calculators are built and tested is described on the Methodology page. If a figure looks wrong, please report it.

Frequently Asked Questions

Why not just A/B test instead?

Shadow mode carries no user risk because outputs are discarded, so it is the right first step for a model whose failure modes are unknown. A/B testing follows once you trust it enough to serve.

What should be compared during a shadow run?

Prediction distributions, latency percentiles, error rates and agreement with the incumbent. Business metrics need an A/B test because shadow outputs never influence users.