Estimation · Estimate

Peak load is 60,000 requests/s. One server handles 2,000 requests/s at most, and you want each to run at no more than 60% of that. How many servers?

easy

Answer

About 50 servers. In review, anything from 33.3 to 75 counts as right.

The worked estimate

Usable capacity per server: 2,000 × 0.6 = 1,200 requests/s. 60,000 ÷ 1,200 = 50 servers. Add a few more so that losing a machine (or a whole zone) still leaves enough.

Why

Running servers near 100% makes queues, and so latency, grow sharply; a target utilization of 50–70% leaves room for spikes and failures.

Numbers: Numbers to know.

Review this in your daily deck All cards in Estimation