Little's Law
Here is a fact so simple it looks like it must have fine print, and so general it borders on the
magical. In any queueing system that has settled into a steady rhythm — a coffee shop, a
motorway, a hospital ward, a warehouse, the packets in a router — the average number of items
inside the system equals the rate at which items arrive, multiplied by the average
time each one spends there. That is Little's law:
L = \lambda\, W.
No assumption about how arrivals are spaced, no assumption about the service-time distribution, no
assumption about the number of servers or the queue discipline. Poisson or not, one server or a
thousand, first-come-first-served or utter chaos — as long as the system is stable and
in the long run, the three quantities are locked together by that single equation.
What the three letters mean
- L — the average number of customers in the system
(waiting plus being served), taken over a long time.
- \lambda — the arrival rate: how many customers enter
per unit time (and, in steady state, the rate at which they leave).
- W — the average time a customer spends in the system,
from arriving to departing.
The units make it click: \lambda is customers per hour,
W is hours, so \lambda W is a pure count
of customers — exactly what L is. Keep your units consistent (per-hour with
hours, per-minute with minutes) and the law never lies.
Why it is true — an accounting argument
Little's law is really just bookkeeping. Picture the number of customers in the system,
N(t), as it rises and falls: up one at each arrival, down one at each
departure. Watch it over a long interval [0, T].
The area under this staircase is the total customer-time accumulated — add up how long
every customer was present. Now count that same area two ways. Averaging height over time
gives area = L \times T. Adding up the horizontal strips, one per customer,
gives area = (\text{customers served}) \times W = (\lambda T)\times W. Two
expressions for one area:
L\,T = \lambda\,T \cdot W \quad\Longrightarrow\quad L = \lambda W.
That is the whole proof. Nothing about randomness entered — which is exactly why the law survives no
matter what the distributions are.
Worked example
A busy sandwich shop serves \lambda = 30 customers per hour, and a customer
spends on average W = 0.1 hours (6 minutes) in the shop from joining the
queue to walking out with lunch. Then, at any random glance,
L = \lambda W = 30 \times 0.1 = 3 \text{ customers in the shop.}
You never counted heads — you deduced the crowd from a rate and a duration. Turn it around: if the
manager wants the shop never to feel more crowded than L = 2, and arrivals
stay at 30/hour, then each customer's time must fall to
W = L/\lambda = 2/30 \approx 0.067 hours — exactly 4 minutes. Little's law
turns a vague "make it feel less busy" into a hard target for the service time.
It applies to any sub-system too
Draw a box around any part of the process and the law holds for that box, as long as it is
stable. Draw the box around just the waiting line (exclude service) and you get the
"queue-only" version:
L_q = \lambda\, W_q,
where L_q is the average number waiting and
W_q the average wait before service. Draw it around the servers
alone and you recover that the average number of busy servers equals
\lambda \times (\text{mean service time}) — the offered load. One law, applied
to nested boxes, ties a whole system together. Below are a few boxes and the number each implies.
| System | Arrival rate λ | Time in system W | Number present L = λW |
|---|
| Sandwich shop | 30 / hour | 0.1 hour | 3 customers |
| A&E department | 8 / hour | 2.5 hours | 20 patients |
| Web server | 500 / second | 0.04 second | 20 requests |
| Warehouse stock | 1200 units / week | 3 weeks | 3600 units on hand |
That last row is a favourite of supply-chain analysts: average inventory equals throughput times the
average time a unit sits in the warehouse. Little's law and inventory theory are the same idea wearing
different hats.
John Little proved the law rigorously in 1961, though the relationship had been folklore among
traffic engineers for years. What Little supplied was the proof that L = \lambda W
holds under breathtakingly mild conditions — essentially just that the system reaches a long-run
steady state and that the averages exist. It does not need the arrivals to be Poisson, does not need
exponential service, does not even need the customers to be served in the order they arrived. This is
why it is one of the most-used results in all of operations research: whenever you know two of the
three quantities, the third comes free.
Little's law is an equation about long-run averages, and it quietly assumes the system is
stable — arrivals are not outpacing departures, so N(t) is
not drifting upward forever. Point it at a shop in its first ten frantic minutes after opening, or at
an overloaded queue that is still growing, and the "averages" have not settled, so the law does not yet
apply. It also says nothing about variability: two systems with identical
L, \lambda and W can
have wildly different worst-case waits. The law pins the means together — it does not describe
the swings around them.