Queue with limited waiting room calculator (M/M/c/K): how many arrivals are turned away?

When a line has a fixed number of waiting spaces, such as a yard, a staging lane or a call queue limit, arrivals that find it full are lost. Enter arrivals, service time, servers and waiting spaces to get the share turned away, the wait of those served and the servers you need for a target.

4 servers, no waiting room20.6%30 arrivals an hour × 6 min = 3.0 servers of work (Erlang B)
Same, 4 waiting spaces4.6%average wait of those served 1.3 min
Same, 8 waiting spaces1.3%average wait 2.3 min: space trades turn-aways for waiting

How to size a queue with a waiting limit

  1. Enter arrivals per hour and the average service time in minutes.
  2. Enter the number of servers and how many customers can wait: 0 if nobody can.
  3. Set the share of arrivals you can afford to turn away.
  4. Read the share lost, the wait of those served and the fewest servers that meet the target.
  5. Press the simulation check, which counts lost arrivals in a seeded run.

Your queue with a limit

Starting values are an example, not data about your site. An arrival that finds every server busy and every waiting space taken is lost and never comes back. The target is your business choice.

Result

Turned away4.58% M/M/c/K blocking
Lost per hour1.38 of 30 arriving
Busy share of servers72%
Average wait, those served1.3 min
Share turned away, 4 servers, by waiting spaces
  • 0 spaces20.6 %
  • 1 spaces13.4 %
  • 2 spaces9.1 %
  • 4 spaces4.6 %
  • 8 spaces1.3 %
  • 16 spaces0.1 %

With 4 waiting spaces, the fewest servers that keep at most 1% turned away is 6.

Open this exact case in Simflowly

Exact M/M/c/K: random arrivals, exponential service, identical servers, K = servers + waiting spaces. Impatient customers who leave the line, appointments and priorities are not modelled.

The formula

a = λ ÷ μ (offered load, in servers of work) pₙ ∝ aⁿ ÷ n! for n = 0 … c pₙ ∝ aⁿ ÷ (c! × c^(n−c)) for n = c+1 … K share turned away = p_K (K = servers + waiting spaces) served per hour = λ × (1 − p_K)

The pₙ are the long-run probabilities of finding n customers in the system, scaled to add up to 1. An arrival is lost when it finds K in the system, and by the PASTA property that share equals p_K. Setting K = c gives the Erlang B formula.

Worked example: 4 servers and a limit

30 arrivals an hour, 6 minutes of service each, so the offered load is 3.0 servers of work on 4 servers. The only thing that changes is how many can wait. Exact formula against the engine (seed 7, 30 replications of 200 hours):

Four servers with 0, 2, 4 and 8 waiting spaces: share turned away and average wait, formula and simulation
Waiting spacesTurned away, formulaTurned away, simulatedAverage wait, formulaAverage wait, simulatedServers busy
020.61%20.79%0 sec0 sec60%
29.13%9.16%40 sec41 sec68%
44.58%4.67%1.3 min1.4 min72%
81.33%1.33%2.3 min2.3 min74%

The first row is Erlang B: 20.6% turned away and nobody waits. Each added space saves arrivals and costs waiting time; the busy share rises because fewer arrivals are lost. To keep at most 1% turned away with 4 spaces you would need 6 servers. Waits are for those served, not for the lost.

When to use this and not the neighbours

  • Nobody can wait (a fully busy line is lost): use the Erlang B calculator.
  • Anyone can wait, with no limit: use the Erlang C calculator.
  • A limited room, such as a yard with room for n trucks, a staging lane of n pallets or a queue cap of n callers: use this page.

When this model is wrong

  • People give up. Customers who join and then leave need an abandonment model, which is not included.
  • Lost means gone. If lost customers retry later, the real load is higher than the one you entered.
  • Service is not exponential. The formula assumes it. With another shape, blocking and waits can differ: run the simulation with your numbers.

Questions

What does M/M/c/K mean?

Random arrivals, exponential service times, c identical servers, and room for K customers in total, servers included. An arrival that finds K in the system is lost. With K equal to c nobody can wait, which is the Erlang B loss system; with K very large it approaches the unlimited Erlang C queue.

How do I size a yard or staging lane?

Treat the lane as the waiting spaces and the doors or stations as servers. Add spaces until the share turned away meets your target, then check the wait of those served, since each added space also adds waiting time. The example shows blocking falling from 20.6% to 1.3% while the wait grows.

Is it cheaper to add a server or a waiting space?

It depends on your costs; the calculator gives the numbers for both. A server cuts both turn-aways and wait. A space only cuts turn-aways and adds wait. In the example, 4 servers with 8 spaces turn away 1.3% but served customers wait 2.3 min; 6 servers with 4 spaces reach the 1% target.

What does this model leave out?

Customers who leave the line after waiting, those who retry later, priorities, appointments and service times that are not exponential. Blocking is exact for exponential service; with another shape it can differ, and the simulation button runs your numbers through the engine to show by how much.

Keep going

Updated 20 November 2026. Free to use, no account. The calculation runs in your browser; nothing you type is sent to us. The engine behind the simulation checks is validated against exact queueing results.