A/B Testing CPU Requirement Calculator FAQ
Answers to common questions about estimating monthly A/B testing CPU core-hours, average capacity, traffic allocation, and overhead.
This FAQ explains the inputs and outputs used in an A/B testing CPU requirement estimate. The calculator is intended for general capacity planning and does not replace production measurement or peak-load analysis.
General CPU planning questions
Core concepts behind the monthly CPU estimate.
What does this A/B testing CPU calculator estimate?
It estimates monthly application CPU core-hours for experiment traffic, including an added percentage for experiment-related processing.
What is a CPU core-hour?
It is the work performed by one CPU core running for one hour. Ten core-hours can be one core for ten hours or ten cores for one hour.
Is the result the CPU capacity I should provision?
It is an average workload estimate. Actual provisioned capacity also depends on peak traffic, resilience needs, utilization targets, and operational headroom.
Does the calculator measure compute cost?
No. It reports CPU workload and average capacity. Pricing depends on the infrastructure, billing model, and resources used.
Inputs and traffic allocation
How to choose request, traffic, and variant inputs.
Should monthly requests include every site request?
Enter requests that are eligible to reach the tested experience. Then use the test traffic percentage for the share actually enrolled.
Should the control count as a variant?
Yes. A normal A/B test has two variants: the control and one treatment.
What traffic percentage should I use for a gradual rollout?
Use the expected average share for the month, or calculate each rollout stage separately and add the resulting core-hours.
What if variants receive unequal traffic?
The total monthly CPU estimate can still be useful, but the per-variant output will not be accurate. Estimate each variant separately if their traffic or CPU behavior differs.
Formula and overhead questions
How request CPU time and experiment overhead affect the result.
How is request CPU time converted to core-hours?
Experiment requests are multiplied by CPU milliseconds per request and divided by 3,600,000 milliseconds per hour.
What should be included in experiment CPU overhead?
Consider assignment logic, feature-flag evaluation, targeting, logging, analytics, metrics, and variant-specific application work not included in baseline CPU time.
Can overhead be zero?
Yes, if the CPU time per request already includes all experiment processing or if additional work is known to be negligible.
Does adding variants automatically raise CPU requirements?
Not in the equal-split formula when total test traffic and request processing are unchanged. More variants may raise CPU if they add distinct code paths, rendering, logging, or service calls.
Accuracy and limitations
What the result excludes and how to validate it.
Why is average CPU lower than CPU needed at peak times?
The average divides total monthly work by all month hours. Real demand may be concentrated into a much smaller number of busy hours.
Does this estimate database and memory usage?
No. It covers application CPU time only. Plan database, cache, memory, network, storage, and queue resources separately.
How can I improve the estimate?
Measure representative CPU time with and without experiment processing, use expected traffic patterns, and compare the estimate with production monitoring.
Can deployments affect actual CPU use?
Yes. Deployments, cache warming, retries, cold starts, and changed code paths can increase CPU beyond the modeled average.
What is a CPU core-hour?
A CPU core-hour is one CPU core used continuously for one hour, or an equivalent amount of CPU work spread across multiple cores.
Explore Related Questions
Ready to see what you can calculate?
Open the calculator and get personalized results in seconds.