Erlang B calculator: how many servers, beds or bays keep blocking low?

Erlang B gives the chance that a new arrival finds every one of c servers busy and is turned away, when there is no waiting room. Enter arrivals per hour and the time each job holds a server, and the calculator returns the fewest servers that keep blocking under your target. It fits beds, bays, dock doors, phone lines and machines.

6 servers, load 6 Erl26.5%simulation 26.7%
8 servers12.2%simulation 12.3%
10 servers4.3%simulation 4.4%

Your loss system

Offered load a = arrivals per hour × hours per job = 6 Erlangs. A job that finds every server busy is lost and does not come back. Starting values are an example; the target is a business choice, not a standard.

Result

Servers needed10 blocking ≤ 5%
Offered load6 Erl
Blocking at 104.31%
Busy share of each server57% carried load ÷ servers
Blocking probability by number of servers, formula and simulation
ServersErlang BLost per hourSimulation
626.49%7.95check
718.51%5.55check
812.19%3.66check
97.51%2.25check
104.31%1.29check
112.3%0.69check
121.14%0.34check

Exact Erlang B. Simulation: random arrivals, exponential service, queueCapacity 0 (nobody waits), 12 seeded runs; it reports lost ÷ arrivals.

Simulate this in Simflowly

What Erlang B answers

Picture c servers and no waiting room. Jobs arrive at random, each one holds a server for some time, and a job that arrives when all c are busy is lost. Erlang B gives the long-run share of arrivals that are lost. It is the formula behind sizing phone trunks, and it applies to anything with that rule: hospital beds when full means diversion, dock doors when trucks are sent away, parking, machines that reject orders.

It is not Erlang C. If arrivals wait instead of leaving, use the queue calculator.

Formula and recursion

a = arrivals per hour × hours per job (offered load, Erlangs) B(a, c) = (a^c / c!) ÷ Σ(k=0..c) a^k / k! B(a, 0) = 1 B(a, c) = a · B(a, c − 1) ÷ (c + a · B(a, c − 1)) (stable recursion)

The recursion is how the calculator computes it: it never forms the large factorials, so it works for hundreds of servers. Blocking falls with every server added, and each added server helps less than the one before.

Worked example

30 arrivals an hour, 12 minutes each: offered load a = 6 Erlangs. With 6 servers, load equals capacity and about a quarter of arrivals are lost. Adding servers helps fast at first:

Erlang B blocking and simulated blocking for 6, 8 and 10 servers at 6 Erlangs
ServersErlang B formulaSimulation
626.5%26.7%
812.2%12.3%
104.3%4.4%

Does a simulation agree?

Yes, and you can check it. The simulation is one source, one station with queue capacity 0 (nobody waits, a full station loses the arrival) and one sink, run with random arrivals and exponential service, seed 7, 30 replications of 200 hours after a 10-hour warm-up. The simulated figure is lost arrivals divided by all arrivals. The formula and the simulation agree to within run-to-run noise. The calculator's check button does the same thing live for your numbers, with a shorter run.

Where it breaks

  • Retries. Callers who try again make arrivals higher than the first-attempt rate; the formula uses the rate you give it.
  • Bunched arrivals. Erlang B assumes random (Poisson) arrivals. Waves and batches usually make blocking worse.
  • Time of day. Demand that peaks in the afternoon needs sizing at the peak, not at the daily average. Simflowly sources take a 24-hour arrival profile, so you can simulate the day.
  • Abandonment and priorities. Callers who give up while waiting, or high-priority arrivals that bump others, are different models and are not simulated by Simflowly.

Questions

What is the difference between Erlang B and Erlang C?

Erlang B has no waiting room: an arrival that finds every server busy is lost, and the answer is the probability of that. Erlang C has an unlimited line: nobody is lost, and the answer is the probability of waiting and the average wait. Use B for beds, parking bays and phone lines that reject callers; use C for service points where people queue.

How do I compute offered load?

Multiply arrivals per hour by the hours each job occupies a server. 30 arrivals an hour at 12 minutes each is 30 × 0.2 = 6 Erlangs: on average six servers are busy if none are ever blocked.

What blocking target is normal?

There is no universal value. One to two percent is a long-standing telephone-network convention, not a law. A hospital, a dock or a parking lot picks its own target from the cost of turning someone away against the cost of an idle server.

Can Erlang B size hospital beds?

Erlang loss models have been used in published research on ward capacity. Real wards have arrivals that change through the day and the week, stays that vary widely, and patients who are held in other units instead of being turned away, so treat the answer as a first cut and not as a staffing or clinical decision.

Does the shape of service time matter?

For the blocking probability, no: Erlang B depends only on the average service time, a known property of the Erlang loss system. That is general theory, not something this page tests. The simulation check uses exponential service times.

Why might real blocking differ from the formula?

Rejected callers or customers who retry, arrivals that are bunched rather than random, and demand that changes by the hour all break the assumptions. Simflowly models an hourly arrival profile; retries and abandonment are not modelled.

Keep going

Updated 17 November 2026. Free to use, no account. The calculation runs in your browser; nothing you type is sent to us.