Little's Law
In a stable system, the average number of items in it equals the arrival rate times the average time each spends in it — L = λW.
The most useful piece of queueing theory for an architect, because it relates the three quantities you actually have numbers for. If a service handles 500 requests per second at 200 ms average latency, it has 100 requests in flight on average, so a thread pool of 50 is a bottleneck and a pool of 5,000 is a memory problem.
Rearranged, it sizes connection pools, worker counts and queue depths, and it explains why latency climbs sharply as utilisation approaches capacity: as λ approaches the service rate, W grows without bound. This is why running a system at 95% utilisation is a decision to have long tail latency.
It assumes stability — arrivals equal departures over the window. When a system is saturating, Little's Law is not violated, it is simply telling you the queue is growing.