Skip to content
Tutorials Center
My learning

Topics

Lesson 15 of 38 · Multi-Server Queues

The M/M/c Model: Several Servers, One Line

Meet the M/M/c queue: c identical servers sharing one line. Derive its steady-state distribution from the birth-death formula and check it numerically.

Many real systems have several servers working from one shared line: a bank with a row of tellers, an airport check-in desk, a pool of web workers. The standard model is the M/M/c queue: Poisson arrivals at rate λ, c identical servers each with exponential service at rate μ, one common FIFO line and unlimited room.

Offered load and stability

Recall the offered load a = λ/μ (in erlangs) and the utilization per server ρ = a/c = λ/(cμ). The system is stable exactly when ρ < 1, that is λ < cμ. Four tellers who each handle 12 customers per hour can sustain fewer than 48 arrivals per hour.

Steady state from birth-death

M/M/c is a birth-death process with λn = λ and death rate μn = min(n, c) μ:

  • With n < c customers present, all of them are in service, so the departure rate is nμ.
  • With n ≥ c, all c servers are busy, so the departure rate is cμ.
flowchart LR
  S0["0"] -- "λ" --> S1["1"]
  S1 -- "μ" --> S0
  S1 -- "λ" --> S2["2"]
  S2 -- "2μ" --> S1
  S2 -- "λ" --> S3["3 = c"]
  S3 -- "3μ" --> S2
  S3 -- "λ" --> S4["4"]
  S4 -- "3μ" --> S3

State diagram for c = 3. The downward rate grows with the number of busy servers, then stays at 3μ.

The product form gives

  • Pn = (a^n / n!) P0 for 0 ≤ n ≤ c
  • Pn = (a^c / c!) ρ^(n-c) P0 for n ≥ c

and normalization fixes

P0 = [ Σ_{k=0}^{c-1} a^k / k! + a^c / (c! (1 - ρ)) ]^(-1)

For c = 1 this collapses to the M/M/1 result P0 = 1 - ρ. Beyond c, the probabilities decay geometrically with ratio ρ, just like M/M/1.

Computing the distribution

The code below builds the distribution from the formulas and also from the generic birth-death solver of the earlier lesson, to confirm they match. We take c = 3, λ = 2.4, μ = 1 (so a = 2.4 and ρ = 0.8).

Python M/M/c steady-state distribution

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The last line demonstrates an identity from Little's law: the average number of busy servers is λ/μ = a, whatever c is (as long as the queue is stable).

Checking stability and load

Before using any M/M/c formula, check λ < cμ. The helper below also reports how much of each server's time is used.

Python Stability check

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The line with ρ = 1 is labeled unstable by the code. At exactly full load, random variation makes the queue grow without a steady state.

Why the formulas look harder than M/M/1

The extra complexity comes from the first c states, where the service rate depends on how many customers are present. The probability that an arriving customer must wait, because all servers are busy, is the key quantity. It has its own name and formula, the Erlang C formula, covered next.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. Six agents each handle 10 calls per hour. What is the largest arrival rate for a stable system?

    Stability requires lambda < c * mu = 6 * 10 = 60.
  2. 2. In an M/M/c queue, what is the death (service completion) rate when n < c customers are present?

    All n customers are in service, each at rate mu, so the total rate is n * mu.
  3. 3. When n >= c, the probabilities P(n) decay by what factor per extra customer?

    For n >= c, P(n+1) = rho * P(n), a geometric decay as in M/M/1.
  4. 4. What is the average number of busy servers in a stable M/M/c queue?

    By Little's law the number in service averages lambda times the mean service time, a = lambda / mu.

Summary

  • M/M/c: Poisson arrivals, exponential service, c servers, one shared line.
  • Stable iff ρ = λ/(cμ) < 1.
  • Death rate is min(n, c) μ; Pn = a^n/n! P0 up to c, then geometric with ratio ρ.
  • The average number of busy servers is a = λ/μ.

Next: The Erlang C Formula