Queue Math: Why a Server That Is 90% Busy Creates Long Waits
October 29, 2026
Waiting lines grow dramatically as a server approaches full utilization. The simplest model (M/M/1) assumes random arrivals at an average rate λ, random service times with an average rate μ, and one server. Utilization is ρ = λ ÷ μ. The average number in the system is L = ρ ÷ (1 − ρ), the average number waiting is Lq = ρ² ÷ (1 − ρ), the average time in the system is W = 1 ÷ (μ − λ), and the average wait in the queue is Wq = ρ ÷ (μ − λ).
Suppose service takes 5 minutes on average, so μ = 0.2 customers per minute. If customers arrive at λ = 0.16 per minute (one every 6.25 minutes), ρ = 0.8, and W = 1 ÷ 0.04 = 25 minutes in total, with 20 minutes in the queue and 5 minutes in service.
Skip the math and get your answer instantly:
Open the Percentage Calculator →The Hockey-Stick Curve
As utilization increases, delay grows slowly at first and then explosively. At ρ = 0.5 (λ = 0.10), the average wait in the queue is 5 minutes. At 0.8, it is 20 minutes, four times as long for 60% more traffic. At 0.9 (λ = 0.18), the queue wait is 45 minutes, and at 0.95 (λ = 0.19), it is 95 minutes, 19 times the service time. At 100% utilization with random arrivals, the queue grows without limit. This is why a system that looks efficient on average can feel terrible to users, and why highways, call centers, and emergency rooms need spare capacity.
| Utilization | Arrivals/min | Avg. in queue (Lq) | Avg. wait in queue (Wq) | Total time (W) |
|---|---|---|---|---|
| 50% | 0.10 | 0.5 | 5 min | 10 min |
| 80% | 0.16 | 3.2 | 20 min | 25 min |
| 90% | 0.18 | 8.1 | 45 min | 50 min |
| 95% | 0.19 | 18.05 | 95 min | 100 min |
Implications for Staffing
Target utilization below about 70% to 85% for systems where waiting matters. Adding a second server doesn't just halve the wait, it can reduce it by far more because the load per server falls; pooling servers into a single shared line is more efficient than separate lines. Reducing the variability of service times also shortens queues: the M/D/1 model with constant service times has half the average queue of M/M/1.
Limits of the Model
Real arrivals are not always random (appointments smooth them; rush hours concentrate them), service times may vary more or less than assumed, and customers may leave if the line is long (balking) or switch lines. For multiple servers, use the M/M/c model (Erlang C). Simulation is a good option for complex cases. Still, the central lesson holds: delay increases sharply as utilization approaches 100%.
Using It
Measure your arrival and service rates for different periods, compute the utilization, and check if the predicted waits match experience. Consider scheduling more servers at peak hours instead of running a constant staff level.
Frequently Asked Questions
What is utilization in a queue?
The arrival rate divided by the service rate.
Why do waits explode near 100% utilization?
Random bursts of arrivals can't be absorbed when there's no spare capacity.
What is the M/M/1 wait in the system?
W = 1 ÷ (μ − λ).
Is one shared line better than separate lines?
Generally yes, because it pools capacity.
How does variability affect queues?
Less variable service times reduce the queue.