A/B Testing CPU Requirement Calculator Examples
Worked examples showing how concurrent users, request rates, CPU time, and utilization affect A/B test CPU capacity.
These examples show how to turn expected A/B test traffic into total CPU core capacity and per-user capacity. They use simplified, steady-state assumptions to illustrate how each input changes the result.
Moderate traffic, two-variant test
A service expects 2,000 concurrent users. Each makes 4 requests per minute, and test processing adds 3 ms to 15 ms of baseline CPU work.
Input Summary
Concurrent users
2,000 users
Requests per user per minute
4 requests/minute
Baseline CPU time
15 ms/request
Experiment overhead
3 ms/request
Variants
2
Target utilization
70%
Calculation Breakdown
- 1Request rate2,000 × 4 ÷ 60133.33 requests/second
- 2CPU time per request15 + 318 ms/request
- 3CPU work133.33 × 18 ÷ 1,0002.40 CPU seconds/second
- 4Required capacity2.40 ÷ 0.703.43 cores
- 5Per-user capacity3.43 ÷ 2,0000.001714 cores/user
Result Summary
Per-user capacity
0.001714 cores/user
A/B Testing CPU Requirement Calculator
The estimate is 3.43 CPU cores, with 1,000 concurrent users expected in each variant under an even split.
High-traffic experiment with conservative headroom
A consumer application expects 50,000 concurrent users making 6 requests per minute. Baseline CPU time is 12 ms and experiment overhead is 4 ms.
Input Summary
Concurrent users
50,000 users
Requests per user per minute
6 requests/minute
Baseline CPU time
12 ms/request
Experiment overhead
4 ms/request
Variants
3
Target utilization
60%
Calculation Breakdown
- 1Request rate50,000 × 6 ÷ 605,000 requests/second
- 2CPU time per request12 + 416 ms/request
- 3CPU work5,000 × 16 ÷ 1,00080 CPU seconds/second
- 4Required capacity80 ÷ 0.60133.33 cores
- 5Users per variant50,000 ÷ 316,667 users per variant
Result Summary
Users per variant
16,667 users per variant
A/B Testing CPU Requirement Calculator
The estimate is 133.33 CPU cores, or 0.002667 cores per concurrent user.
Low-traffic test with heavier variant processing
A B2B application has 500 concurrent users making 10 requests per minute. Baseline CPU time is 30 ms and experiment overhead is 20 ms.
Input Summary
Concurrent users
500 users
Requests per user per minute
10 requests/minute
Baseline CPU time
30 ms/request
Experiment overhead
20 ms/request
Variants
4
Target utilization
75%
Calculation Breakdown
- 1Request rate500 × 10 ÷ 6083.33 requests/second
- 2CPU time per request30 + 2050 ms/request
- 3CPU work83.33 × 50 ÷ 1,0004.17 CPU seconds/second
- 4Required capacity4.17 ÷ 0.755.56 cores
- 5Users per variant500 ÷ 4125 users per variant
Result Summary
Users per variant
125 users per variant
A/B Testing CPU Requirement Calculator
The estimate is 5.56 CPU cores, or 0.011111 cores per concurrent user.
How to Read Your Results
Total CPU cores are the estimated application CPU capacity needed at the selected average utilization target.
CPU cores per user is a normalized measure for comparing workload efficiency across traffic levels.
Requests per second reflects concurrent users and their request frequency, not total registered users.
Users per variant assumes an even traffic split and helps assess the expected scale of each treatment.
Treat decimal core results as workload estimates; actual deployment capacity may need to align with available instance or container sizes.
Assumptions & Important Notes
- Each scenario assumes a stable request rate throughout the measurement period.
- CPU times are average CPU processing times for representative requests.
- Variants are evenly split for the per-variant user result.
- No separate reserve is added for operating system work, background processing, failures, or non-application services.
Related Examples
Frequently Asked Questions
Can I use these examples for container CPU limits?
They can provide a starting workload estimate, but container limits and requests also depend on scheduling, burst behavior, throttling, and the deployment environment.
Why can a low-traffic example need more CPU per user?
CPU per user rises when each request is more CPU-intensive, users send requests more often, or a lower utilization target is selected.
What happens if I double requests per user per minute?
With other inputs unchanged, request rate, CPU work, total required cores, and per-user CPU capacity all double.
Do four variants divide total CPU by four?
No. Variants divide users per variant under an even split. Total CPU stays tied to total request volume and CPU time per request.
Ready to calculate your own result?
Use the live calculator with your own inputs, timing, and preferences.