Little's Law
· performance, queueing
For any stable queueing system, L = λW:
- L — average number of items in the system
- λ — average arrival rate
- W — average time an item spends in the system
It holds regardless of arrival distribution, service distribution, or scheduling discipline, which is what makes it useful — you can measure any two and derive the third without modelling the system.
Practical use: concurrency limit and target latency give you a throughput ceiling. A pool of 48 connections at a 20ms mean service time cannot exceed 2400 requests per second, no matter how the load is shaped. If you are asking for more than that, no amount of tuning below the pool will produce it.
Related: Percentiles lie under load.
Link into a private note, which must stay unresolved and must not reveal its title: private-scratch.