M/M/c queue example, solved step by step: three servers at 80% utilization

Customers arrive at 2.4 a minute, each server serves 1 a minute and there are 3 servers. The Erlang C formula gives a 64.7% chance of waiting and an average wait of 1.0787 minutes in line. The engine, run 30 times, agrees: 1.0850 minutes, with the exact value inside its 95% interval.

Probability of waiting C(3, 2.4)64.72%simulated 64.80%
Average wait in line Wq1.0787 minsimulated 1.0850 (1.0723 to 1.0978)
Average number waiting Lq2.5888simulated 2.6052

The problem

A service point receives customers at random at λ = 2.4 per minute. Each of c = 3 identical servers handles μ = 1 customer per minute, with exponential service times, and all share one line. Find the probability that a customer waits, the average wait, the average line length and the average time in the system. This is the standard M/M/c exercise; it uses the same numbers as the M/M/3 rows of the engine validation.

Step 1: offered load and utilization

a = λ ÷ μ = 2.4 ÷ 1 = 2.4 (servers of work) ρ = a ÷ c = 2.4 ÷ 3 = 0.7999999999999999 (must be below 1)

Step 2: the Erlang C probability of waiting

C(c, a) = [aᶜ ÷ (c! × (1 − ρ))] ÷ [ Σₖ₌₀..c−1 aᵏ ÷ k! + aᶜ ÷ (c! × (1 − ρ)) ]
Terms of the Erlang C sum for a = 2.4 and c = 3
kTerm aᵏ ÷ k!Value
02.4^0 ÷ 0!1.000
12.4^1 ÷ 1!2.400
22.4^2 ÷ 2!2.880
Sum for k = 0 to 26.280
Last term: 2.4^3 ÷ (3! × (1 − 0.7999999999999999)) = 13.824 ÷ (6 × 0.2)11.520

C = 11.520 ÷ (6.280 + 11.520) = 0.6472. About 64.7% of customers find all three servers busy and have to wait.

Step 3: the averages

Wq = C ÷ (c·μ − λ) = 0.6472 ÷ (3 × 1 − 2.4) = 1.0787 min Lq = λ × Wq = 2.4 × 1.0787 = 2.5888 W = Wq + 1 ÷ μ = 1.0787 + 1 = 2.0787 min L = λ × W = 2.4 × 2.0787 = 4.9888

Step 4: check it with the engine

The same system built in the Simflowly engine (exponential arrivals and service, seed 2026, 30 replications of 100,000 simulated minutes, 5,000-minute warm-up) gives:

Exact values and simulated means for the M/M/3 example
MeasureExactSimulated mean95% interval
Utilization0.80000.80070.7997 to 0.8018
P(wait > 0)0.64720.64800.6462 to 0.6498
Wq (min)1.07871.08501.0723 to 1.0978
Lq2.58882.60522.5732 to 2.6372
W (min)2.07872.08562.0724 to 2.0988

Every exact value lies inside its 95% interval. These are the engine's rows in the validation table; running the script again reproduces them because the seed is fixed.

Step 5: what if you add a server?

Same arrivals and service rate with 2 to 5 servers
Servers cρP(wait > 0)Wq
21.201unstable
30.800.64721.1 min
40.600.287011 sec
50.480.11353 sec

With 2 servers the system is unstable (ρ = 1.2). The fourth server cuts the wait from 1.1 min to 11 sec, a reduction of 83%, much more than the 25% more capacity might suggest, because the system leaves 80% utilization where the line grows quickly. To try other numbers, use the queue calculator.

Questions

How do you solve an M/M/c queue by hand?

Find the offered load a = λ ÷ μ and the utilization ρ = a ÷ c. Evaluate the Erlang C probability of waiting from the sum of aᵏ ÷ k! terms. Then compute Wq, Lq, W and L from it, using Little’s law for the lengths. The steps below show every number.

What is the difference between Wq and W?

Wq is the average time in line before service starts, 1.0787 minutes here. W is the average time in the system, wait plus service: 2.0787 minutes, since service takes 1 minute on average. Lq counts only customers waiting, 2.5888, while L also counts those being served, 4.9888.

Is the answer an average or a guarantee?

An average over a long run. The chance that a customer waits is 64.7%, so about one in three does not wait at all, while those who wait wait longer than the average suggests. Exponential service is also assumed; more regular service times give shorter waits, as the Erlang C vs simulation guide shows.

Keep going