Skip to content
Tutorials Center
My learning

Topics

Lesson 22 of 38 · Finite Capacity and Priorities

Erlang B and Loss Systems

Learn the M/M/c/c loss model, the Erlang B formula and its stable recurrence, and how to find how many lines or channels you need for a blocking target.

Some systems have no waiting room at all. A telephone trunk group with c lines either connects your call or gives a busy signal. A car park with c spaces turns cars away when full. A cell tower with c radio channels blocks new calls when every channel is taken. These are loss systems, and the model for them is the M/M/c/c queue: Poisson arrivals at rate λ, exponential service with rate μ per server, c servers, and capacity c (nobody waits).

A century ago the Danish engineer A. K. Erlang worked out the blocking probability for exactly this problem while planning telephone exchanges. It is still used for trunk and channel sizing today.

Offered load and the formula

The offered load is a = λ/μ, measured in erlangs. It is the average number of servers that would be busy if no call were ever blocked. The state n (calls in progress, 0..c) is a birth-death chain with arrival rate λ and death rate nμ, so p_n ∝ a^n / n!. Normalizing gives the Erlang B formula:

Text
B(c, a) = (a^c / c!) / Σ_{k=0}^{c} (a^k / k!)

B(c, a) is the probability that an arriving call is blocked. Remarkably, the formula holds for any service-time distribution with mean 1/μ (this is called insensitivity), not only the exponential.

A stable way to compute it

Factorials and powers overflow quickly. The recurrence below avoids that:

Text
B(0, a) = 1
B(c, a) = a · B(c-1, a) / (c + a · B(c-1, a))
Python Erlang B by recurrence and by the direct formula

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.

Both columns agree. With a = 10 erlangs and exactly 10 lines, roughly one call in five is blocked. Adding lines helps, but with diminishing returns.

Carried load and utilization

Not every offered call is carried. The carried load is a(1 - B), the mean number of busy servers. Per-server utilization is a(1 - B)/c.

This reveals the central trade-off of loss systems: very low blocking needs many servers that sit idle a lot of the time. Small systems are inefficient; big systems pool their capacity and run hotter at the same blocking level.

Sizing: how many lines do I need?

The design question is usually the inverse: given the offered load a and a blocking target (say 1%), find the smallest c with B(c, a) ≤ target.

Python Find the smallest number of lines for a blocking target

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.

Look at the last column: utilization climbs as the system grows, even though the blocking target is the same. This is the economy of scale in queueing.

Erlang B is for lost calls. If blocked callers wait instead, use the Erlang C formula for the delay system from the multi-server chapter. The two formulas are linked: C = B / (1 - ρ(1 - B)) with ρ = a/c.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. A trunk group is offered 6 erlangs of traffic. What does that number mean?

    Offered load a = λ/μ is the mean number of busy servers in an infinite-server system.
  2. 2. What is B(1, a)?

    With one server the recurrence gives B = a·1/(1 + a·1) = a/(1+a).
  3. 3. If a loss system with c servers has blocking probability B, what is the mean number of busy servers?

    Carried load is the offered load times the fraction accepted.
  4. 4. Why is the recurrence used instead of the direct formula for large c?

    The recurrence only multiplies and divides moderate numbers, so it is numerically stable.

Summary

  • A loss system M/M/c/c has no waiting room; blocked customers leave.
  • Offered load a = λ/μ is measured in erlangs.
  • B(c, a) = (a^c/c!) / Σ a^k/k!, computed stably by B(c) = aB(c-1)/(c + aB(c-1)).
  • Carried load is a(1 - B); larger systems reach higher utilization at the same blocking.
  • Invert the formula by searching for the smallest c meeting the target.

Next: The Machine Repair Problem