A/B Testing CPU Requirement Calculator Examples
Worked examples showing how experiment traffic, request CPU time, and overhead affect monthly A/B testing CPU requirements.
These examples show the difference between low-volume, high-volume, and multi-variant experiment workloads. Each result is a monthly CPU planning estimate and should be compared with observed peak demand before infrastructure changes.
Low-volume feature experiment
A product team tests a new checkout message on 20% of 250,000 eligible monthly requests.
Input Summary
Monthly eligible requests
250,000 requests
Test traffic
20%
Variants
2
CPU time per request
15 ms
Experiment overhead
8%
Planning month
730 hours
Calculation Breakdown
- 1Experiment requests250,000 × 20%50,000 requests
- 2Base CPU50,000 × 15 ÷ 3,600,0000.21 core-hours
- 3CPU with overhead0.21 × 1.080.23 core-hours
- 4Average capacity0.23 ÷ 7300.0003 CPU cores
Result Summary
Average capacity
0.0003 CPU cores
A/B Testing CPU Requirement Calculator
Estimated test CPU use is 0.23 core-hours per month.
Standard two-variant web test
A website routes 50% of one million monthly eligible requests into a two-variant experiment.
Input Summary
Monthly eligible requests
1,000,000 requests
Test traffic
50%
Variants
2
CPU time per request
25 ms
Experiment overhead
10%
Planning month
730 hours
Calculation Breakdown
- 1Experiment requests1,000,000 × 50%500,000 requests
- 2Base CPU500,000 × 25 ÷ 3,600,0003.47 core-hours
- 3Extra experiment CPU3.47 × 10%0.35 core-hours
- 4Monthly CPU and average capacity3.82 ÷ 7303.82 core-hours and 0.005 CPU cores
Result Summary
Monthly CPU and average capacity
3.82 core-hours and 0.005 CPU cores
A/B Testing CPU Requirement Calculator
The test requires about 3.82 core-hours per month, including 0.35 core-hours of experiment overhead.
High-volume multi-variant API experiment
An API team tests four response strategies on 75% of 80 million eligible monthly requests.
Input Summary
Monthly eligible requests
80,000,000 requests
Test traffic
75%
Variants
4
CPU time per request
40 ms
Experiment overhead
15%
Planning month
744 hours
Calculation Breakdown
- 1Experiment requests80,000,000 × 75%60,000,000 requests
- 2Base CPU60,000,000 × 40 ÷ 3,600,000666.67 core-hours
- 3CPU with overhead666.67 × 1.15766.67 core-hours
- 4Average and per-variant CPU766.67 ÷ 744; 766.67 ÷ 41.031 CPU cores and 191.67 core-hours per variant
Result Summary
Average and per-variant CPU
1.031 CPU cores and 191.67 core-hours per variant
A/B Testing CPU Requirement Calculator
The four-variant API experiment is estimated at 766.67 core-hours monthly and 1.031 average CPU cores.
How to Read Your Results
Monthly CPU core-hours measure total CPU work for requests assigned to the experiment during the month.
Average CPU cores spread that work across every planning-month hour; they are not a peak-load estimate.
Extra CPU from testing isolates the overhead above normal request processing for test traffic.
CPU per variant is an even-split allocation, useful for rough comparison rather than precise service-level sizing.
Use observed CPU utilization and traffic peaks to add operational headroom outside this estimate.
Assumptions & Important Notes
- Each example treats CPU milliseconds as application CPU time rather than end-to-end response latency.
- Experiment overhead is applied after calculating base CPU for experiment traffic.
- Traffic is assumed to be evenly spread when calculating average CPU cores.
- Per-variant results assume equal traffic allocation and similar execution costs.
Related Examples
Frequently Asked Questions
Can I use these examples for a feature-flag rollout?
Yes, if the rollout has a known request share, baseline CPU time, and estimated additional flag or logging overhead. Adjust the inputs to match the rollout.
Why can a high-traffic test still show a low average core count?
A core-hour total is spread across all hours in the month. Short traffic peaks can require much more CPU than the monthly average.
What if test traffic changes during the month?
Use a weighted expected traffic share, or calculate separate periods and add their CPU core-hours.
Can I compare different experiments with core-hours?
Yes. Core-hours provide a common workload measure, but comparisons are strongest when CPU time and overhead were measured using comparable conditions.
Ready to calculate your own result?
Use the live calculator with your own inputs, timing, and preferences.