Work out load based instance count instantly with clear inputs, formula shown and shareable results.
Target-tracking autoscaling computes desired capacity as the aggregate metric divided by the per-instance target, which is exactly how AWS and Kubernetes horizontal autoscalers behave. Step size then determines how many scaling actions are needed, and each action carries a cooldown — so a large gap takes several minutes to close regardless of the arithmetic.
Target tracking
desired = ceil(aggregate metric / target per instance); steps = ceil(|desired - current| / step size)
Request count or concurrency per instance, because they respond linearly. CPU is noisier and lags the actual load change.
They prevent oscillation. Without one, the scaler reacts to metrics that have not yet reflected the last change and overshoots in both directions.