5 / 7 Back-of-the-Envelope Estimation

Count The Boxes

Peak traffic is 30,000 QPS. Load tests show one instance sustains 1,000 QPS while staying inside your p99 latency budget. You target 70 percent steady-state utilization and want two spare instances for failures. How many instances should you plan for?

A

About 300, since a 10x buffer over the measured peak absorbs any future spike

B

About 45, since 30 instances carry peak, dividing by 0.7 gives 43, and two spares round it up

C

Exactly 30, since peak QPS divided by per-instance throughput is the provisioning answer

D

About 21, since running 30 instances at a 70 percent utilization target means 21 are needed