L = λ · W

L is the number of requests in the system — being served plus queued. λ is the arrival rate. W is the time a request spends in the system.

It holds for any stable system, with no assumption about the arrival distribution. That is why it is the one queueing result worth memorising.

Read it three ways.

Forwards — 1200 rps at a 250 ms target needs 300 concurrent slots. Backwards — 32 workers at 20 ms of service each is 32/0.020 = 1600 rps of capacity, full stop; no tuning gets more without changing one of those two numbers. Sideways — a dashboard showing 40 in-flight, 1200 rps and 20 ms latency is showing you a wrong number, because 1200 · 0.020 is 24.