A/B Testing CPU Requirement (Monthly) Calculator
Estimate the monthly CPU capacity needed to serve traffic in an A/B test, including experiment-processing overhead.
Overview
This A/B testing CPU requirement calculator estimates the monthly compute needed for traffic included in an experiment. Enter your eligible request volume, the share placed in the test, average CPU time per request, number of variants, and the additional processing overhead caused by experimentation.
How it works
The calculator first estimates requests assigned to the experiment by multiplying monthly requests by the test traffic percentage. It converts average CPU milliseconds per request into core-hours, then increases that amount by your experiment overhead percentage. Dividing monthly core-hours by the hours in the planning month gives the average CPU cores needed if demand is even. The per-variant figure assumes an equal traffic allocation.
How to use this calculator
- 1Enter the monthly number of requests that can reach the tested experience.
- 2Set the percentage of eligible traffic that will participate in the experiment.
- 3Enter the number of variants, including the control.
- 4Add the average CPU time consumed by each request.
- 5Estimate the percentage of extra CPU work created by experimentation.
- 6Review monthly core-hours and average CPU capacity, then allow headroom for peaks.
Example Calculation
Monthly requests
1000000
Traffic included in test
50%
Number of variants
2
Average CPU time per request
25
Experiment CPU overhead
10%
Hours in planning month
730
Monthly CPU requirement
3.82 core-hours
With 500,000 requests in the test, 25 ms of CPU time per request, and 10% experiment overhead, the test requires about 3.82 core-hours per month, or roughly 0.005 average CPU cores.
Frequently asked questions
What is a CPU core-hour?
A CPU core-hour is one CPU core used continuously for one hour. For example, 10 core-hours could mean one core for 10 hours or 10 cores for one hour.
Should I use total site traffic or experiment traffic?
Enter eligible monthly requests, then use the traffic included in test field to specify the portion actually assigned to the experiment.
Does adding more variants always increase total CPU requirements?
Not necessarily. When total experiment traffic stays the same and variants have similar processing, it is mainly divided among more variants. Extra variant logic or separate services may create additional overhead.
Why is average CPU capacity lower than peak capacity?
Average capacity spreads total work across every hour in the month. Real traffic often has daily or campaign-driven spikes, so infrastructure should also be sized for peak demand and headroom.
How can I estimate experiment CPU overhead?
Compare CPU time with and without experiment assignment, feature flags, logging, analytics, and variant logic in a representative environment. If uncertain, use a cautious estimate and monitor actual usage.
Does this calculator include database and memory requirements?
No. It estimates application CPU time only. Database queries, cache behavior, memory, network use, storage, and queue processing should be planned separately.
Explore Related Calculators
Assumptions and warnings
Assumptions
- Traffic is spread evenly throughout the month when converting core-hours into average CPU cores.
- All variants receive an equal share of the experiment traffic.
- The average CPU time per request represents normal processing before experiment-specific work.
- Experiment overhead covers additional targeting, assignment, flag evaluation, logging, analytics, and variant-processing work.
- Results are capacity-planning estimates and do not include memory, network, storage, database, or burst-capacity requirements.
Warnings
- Actual CPU demand can be higher during traffic peaks, deployments, cache misses, retries, or database-intensive requests.
- Leave operational headroom and validate estimates with production monitoring before changing infrastructure capacity.