Compute the desired Auto Scaling capacity from a target-tracking policy and see how long a full scale-out takes.
Target tracking behaves like a thermostat: it solves for the instance count that puts the chosen metric back on its set point, which is load divided by the usable capacity per instance. The scale-out time matters just as much — each activity adds a step of instances and then waits out the warm-up before the next evaluation, so a large gap between minimum and desired is crossed in several slow rounds rather than one jump. Setting a target too high leaves no absorption for the warm-up window, so traffic that arrives faster than instances boot causes errors even though the policy is technically correct.
EC2 Auto Scaling
desired = ceil(load ÷ (capacity per instance × target utilisation)), clamped to the group's minimum and maximum.
desired = ceil(load ÷ (capacity per instance × target utilisation)), clamped to the group's minimum and maximum. Target tracking behaves like a thermostat: it solves for the instance count that puts the chosen metric back on its set point, which is load divided by the usable capacity per instance. The scale-out time matters just as much — each activity adds a step of instances and then waits out the warm-up before the next evaluation, so a large gap between minimum and desired is crossed in several slow rounds rather than one jump.
Setting a target too high leaves no absorption for the warm-up window, so traffic that arrives faster than instances boot causes errors even though the policy is technically correct.
This calculator takes 7 inputs: Offered load, Capacity of one instance at 100%, Target tracking utilisation, Minimum group size, Maximum group size, Instance warm-up, Instances added per scaling activity. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
CloudWatch alarms for target tracking evaluate on one-minute metrics, so even after the warm-up expires the group waits for the next evaluation period before it acts again. The extra minute approximates that detection lag.