Work out auto scaling instance count instantly with clear inputs, formula shown and shareable results.
Autoscaling bounds are set from the baseline and peak request rates divided by per-instance capacity at a safe utilisation of around 70 percent. The saving comes from the shape of the day: a five-hour peak means average running capacity is far below the maximum, and provisioning for peak all day wastes the difference.
Scaling bounds
min = ceil(baseline / (per-instance x 0.7)); max = ceil(peak / (per-instance x 0.7)); average = min + (max - min) x peak hours / 24
Instances should not run at their absolute ceiling. Reserving 30 percent covers request bursts and gives the scaler time to add capacity before latency degrades.
Usually yes, at least two instances across two availability zones, so a single instance or zone failure does not take the service down.