Skip to content

Operations Management and Strategic Management · Scheduling and Queuing Models

Multi-Channel Queuing Model (M/M/s) Explained

Updated 10 October 2026 · Fact-checked

The multi-channel queuing model M/M/s describes a queue with Poisson arrivals, exponential service times and s identical servers sharing one waiting line. You find ρ = λ ÷ (sμ), check ρ < 1, calculate P0, then Lq, Wq, Ws and Ls. For cost questions, compare total cost for different values of s.

Understand Multi-Channel Queuing Model (M/M/s)

A queue forms when customers arrive faster than they can be served for a while. In the single-channel model (M/M/1) one server handles everyone. In the multi-channel model (M/M/s) there are s identical servers, such as 3 counters at a bank, and customers wait in one common line. The first free server takes the next customer.

The letters mean: first M = arrivals follow a Poisson process (random, at an average rate λ); second M = service times are exponential (average rate μ per server); s = number of servers. Other usual assumptions: first come first served, infinite population, no balking or reneging, and all servers work at the same rate.

The key idea is that the system's total capacity is sμ, not μ. The utilisation factor is ρ = λ ÷ (sμ). It must be below 1. If ρ is 1 or more, the queue grows without limit and the steady-state formulas do not apply. So the smallest possible s is the first whole number above λ ÷ μ.

Adding a server cuts waiting but adds service cost. Waiting cost (lost sales, idle staff, unhappy customers) falls as s rises. Service cost rises with s. The best s is where total cost = service cost + waiting cost is lowest. Exam questions often ask you to find it by trying s = 2, 3, and so on.

Compared with M/M/1: a common line with s servers gives a shorter wait than s separate single-server lines with the same total arrivals and capacity, because no server sits idle while someone waits elsewhere.

Key rules to remember

Utilisation factor
ρ = λ ÷ (sμ)
λ = arrival rate, μ = service rate per server, s = servers. Steady state needs ρ < 1.
Traffic intensity (offered load)
r = λ ÷ μ
Average number of servers' worth of work arriving. Also written as the average number being served.
Probability of an empty system
P0 = 1 ÷ [ Σ (n = 0 to s−1) of rⁿ ÷ n! + rˢ ÷ (s! × (1 − ρ)) ]
Compute the sum term by term for n = 0 up to s−1, then add the last term.
Probability that an arrival has to wait
P(wait) = [rˢ ÷ (s! × (1 − ρ))] × P0
This is the last term inside the P0 bracket multiplied by P0.
Average number waiting in queue
Lq = P0 × rˢ × ρ ÷ [ s! × (1 − ρ)² ]
Counts only those waiting, not those being served.
Average waiting time in queue
Wq = Lq ÷ λ
Little's law. Keep the time unit of λ.
Average time in system
Ws = Wq + 1 ÷ μ
Waiting time plus average service time.
Average number in system
Ls = Lq + λ ÷ μ = λ × Ws
Use this as a check on your working.
Total cost per hour
TC = s × Cs + Cw × (Ls or Lq)
Cs = cost per server per hour, Cw = waiting cost per customer per hour. Use Ls if the cost applies to customers in the system, Lq if only to those waiting. Follow the question.

How to solve Multi-Channel Queuing Model (M/M/s) questions

Use this order for any M/M/s numerical. Keep λ and μ in the same time unit throughout.

  1. 1Write down λ, μ (per server) and s. Convert to the same unit, for example per hour.
  2. 2Calculate r = λ ÷ μ and ρ = λ ÷ (sμ). If ρ ≥ 1, state that the queue is unstable and more servers are needed.
  3. 3Calculate P0: add rⁿ ÷ n! for n = 0 to s−1, add rˢ ÷ (s!(1 − ρ)), then take the reciprocal.
  4. 4Calculate Lq = P0 × rˢ × ρ ÷ [s!(1 − ρ)²].
  5. 5Calculate Wq = Lq ÷ λ, Ws = Wq + 1 ÷ μ and Ls = Lq + r. Check that Ls = λ × Ws.
  6. 6If asked, calculate P(wait) as the last term of the P0 bracket multiplied by P0.
  7. 7For a cost question, repeat steps 2 to 5 for each s, compute total cost, and choose the lowest. State the conclusion in a line.
  8. 8Convert time answers to minutes if the question asks, and give the interpretation in words.

Quickest way: Fast route for s = 2 and s = 3

When to use it: When time is short and the question gives s = 2 or 3, or asks you to compare small values of s.

  1. Find r = λ ÷ μ first. Everything else uses r and ρ.
  2. Write the P0 bracket directly: for s = 2 it is 1 + r + r² ÷ (2(1 − ρ)); for s = 3 it is 1 + r + r²÷2 + r³ ÷ (6(1 − ρ)).
  3. Keep fractions rather than rounding early. Round only at the end.
  4. Get Lq once, then find Wq, Ws and Ls by the short relations, not by fresh formulas.
  5. In a cost comparison, calculate Ls once per s and plug into the total cost line. Stop when cost starts rising.

Common mistakes in Multi-Channel Queuing Model (M/M/s)

  • Using the total service rate sμ as μ in the formula for Ws, giving Ws = Wq + 1 ÷ (sμ).

    Students confuse system capacity with the speed of one server.

    Fix: A customer is served by one server, so add 1 ÷ μ, the service time at a single server.

  • Skipping the check ρ < 1.

    Students jump straight to P0.

    Fix: Compute ρ = λ ÷ (sμ) first. If it is 1 or more, no steady state exists and the formulas give nonsense.

  • Leaving out the last term or dividing it incorrectly when calculating P0.

    The bracket has two different parts: a sum up to s−1 and a special final term.

    Fix: Write the sum terms and the final term on separate lines. The last term is rˢ ÷ (s!(1 − ρ)).

  • Mixing time units, for example λ per hour and μ per minute.

    Questions often give arrival per hour and service time in minutes.

    Fix: Convert service time to a rate first. A service time of 10 minutes means μ = 6 per hour.

  • Using Lq in the waiting cost when the question says cost of customers in the system, or the reverse.

    Students do not read what the waiting cost is charged on.

    Fix: Underline the wording. 'Time spent in the system' means Ls. 'Waiting in the queue' means Lq.

  • Treating s separate M/M/1 lines as the same as one M/M/s line.

    Both have the same total capacity.

    Fix: In a common line no server sits idle while someone waits, so the average wait is lower. Say this in theory answers.

Worked examples

Example 1

A clinic has 2 doctors. Patients arrive at an average of 6 per hour (Poisson). Each doctor serves at an average of 4 patients per hour (exponential). Find (a) P0, (b) the probability that a patient has to wait, (c) Lq and Wq, (d) Ws and Ls.

Show the solution
  1. λ = 6 per hour, μ = 4 per hour, s = 2.
  2. r = 6 ÷ 4 = 1.5. ρ = 6 ÷ (2 × 4) = 0.75, which is below 1, so the system is stable.
  3. P0 bracket: n = 0 gives 1; n = 1 gives 1.5. Last term = 1.5² ÷ (2 × 0.25) = 2.25 ÷ 0.5 = 4.5. Total = 1 + 1.5 + 4.5 = 7. So P0 = 1/7 = 0.1429.
  4. P(wait) = 4.5 × (1/7) = 0.6429.
  5. Lq = (1/7) × 2.25 × 0.75 ÷ [2 × (0.25)²] = (1/7) × 1.6875 ÷ 0.125 = (1/7) × 13.5 = 1.9286 patients.
  6. Wq = 1.9286 ÷ 6 = 0.3214 hour, about 19.3 minutes.
  7. Ws = 0.3214 + 1/4 = 0.5714 hour, about 34.3 minutes.
  8. Ls = 1.9286 + 1.5 = 3.4286. Check: λ × Ws = 6 × 0.5714 = 3.4286.

Answer: P0 = 0.1429; P(wait) = 0.6429; Lq = 1.93 patients; Wq ≈ 19.3 minutes; Ws ≈ 34.3 minutes; Ls = 3.43 patients.

Example 2

Customers arrive at a service counter at 10 per hour (Poisson). Each server serves 6 customers per hour (exponential). A server costs ₹200 per hour. The waiting cost is ₹150 per hour for each customer in the system. Compare 2 servers with 3 servers and decide how many to employ.

Show the solution
  1. λ = 10, μ = 6, r = 10 ÷ 6 = 1.6667. One server is not enough, since λ > μ, so start with s = 2.
  2. For s = 2: ρ = 10 ÷ 12 = 0.8333. P0 bracket = 1 + 1.6667 + 1.6667² ÷ (2 × 0.1667) = 1 + 1.6667 + 8.3333 = 11. P0 = 1/11.
  3. Lq = (1/11) × 2.7778 × 0.8333 ÷ [2 × (0.1667)²] = (1/11) × 2.3148 ÷ 0.05556 = (1/11) × 41.667 = 3.788.
  4. Ls = 3.788 + 1.6667 = 5.4545.
  5. Total cost for s = 2 = 2 × 200 + 150 × 5.4545 = 400 + 818.18 = ₹1,218.18 per hour.
  6. For s = 3: ρ = 10 ÷ 18 = 0.5556. P0 bracket = 1 + 1.6667 + 1.3889 + 1.6667³ ÷ (6 × 0.4444) = 1 + 1.6667 + 1.3889 + 1.7361 = 5.7917. P0 = 0.1727.
  7. Lq = 0.1727 × 4.6296 × 0.5556 ÷ [6 × (0.4444)²] = 0.1727 × 2.1701 = 0.3747.
  8. Ls = 0.3747 + 1.6667 = 2.0414.
  9. Total cost for s = 3 = 3 × 200 + 150 × 2.0414 = 600 + 306.21 = ₹906.21 per hour.
  10. Compare: ₹906.21 is less than ₹1,218.18.

Answer: Employ 3 servers. Total cost is about ₹906 per hour against about ₹1,218 per hour with 2 servers. The extra server costs ₹200, but it saves about ₹512 in waiting cost.

Exam tips

  • Write the formula, then substitute, then the answer. Step marks go to the substitution lines, especially P0.
  • In cost questions, find the smallest s with ρ < 1 and test that value and the next one or two. Stop once total cost rises and state the conclusion.
  • Check Ls = λ × Ws before moving on. It catches most arithmetic slips.
  • Read the MCQ carefully: it may ask only for ρ or for the minimum number of servers. For these you do not need P0.
  • In theory questions, list the assumptions of M/M/s and the difference from M/M/1: s servers, one common queue, stability condition ρ = λ ÷ (sμ) < 1.

Practice questions from Scheduling and Queuing Models

Multi-Channel Queuing Model (M/M/s): frequently asked questions

What is the difference between single-channel and multi-channel queuing?

A single-channel model has one server, and stability needs λ < μ. A multi-channel model has s identical servers sharing one queue, and stability needs λ < sμ. With more servers the waiting time falls, but service cost rises.

Why must ρ be less than 1 in the M/M/s model?

ρ = λ ÷ (sμ) compares arrivals with total service capacity. If ρ is 1 or more, customers arrive at least as fast as they can be served, so the queue keeps growing. Steady-state formulas then do not exist.

How do I find the best number of servers?

Start with the smallest s that gives ρ < 1. Calculate total cost per hour (server cost plus waiting cost) for s, s + 1 and so on. Choose the s with the lowest total cost.

Do I need to memorise the P0 formula?

Yes, for numericals with s = 2 or 3 you should be able to write it quickly. Practise writing the bracket as a sum of terms up to s−1 plus the final term rˢ ÷ (s!(1 − ρ)). Then take its reciprocal.