Capacity Model

How much you have, how much of it you can actually plan against, and what is left when something fails. The reserve is the whole exercise: capacity that is fully committed at steady state has no room for an instance failure, a deploy and a burst — which is a combination that occurs far more often than independence would suggest.

SIMULATED

Every number below is arithmetic on the numbers you entered. “Requests one instance can serve safely” is the hard one in real life and it is an input here, not a result: it depends on the workload, the runtime and what the instance is actually contending for. Measure it; do not inherit it from a slide.

What that gives you

Nominal capacity
960 rps

Everything healthy, nothing held back. The number that flatters you.

Usable capacity
720 rps

Excluding the reserve. This is the number to plan against.

Utilisation of usable
83%

Demand against usable capacity. Above 100% means the reserve is already being spent.

Instances you can lose
3

Before nominal capacity stops covering demand. Zero means the next failure is an incident.

Usable capacity covers demand

At the stated demand you are serving from usable capacity, with the reserve still intact for a failure, a deploy or a burst.

Could this fleet absorb a partner region?

If two active regions each normally carry half the traffic, either one must be able to carry all of it. That is a doubling, and it is the part of multi-region that gets budgeted last.

Does not survive region failover

This fleet cannot absorb an equal partner region's traffic. Two active regions at this size are not a failover plan — they are two regions that both fail when either does.