Erlang B vs Erlang C: which one do you need?
Use Erlang B when a customer who finds every server busy is lost and does not wait; use Erlang C when that customer waits in a line without limit. For the same load of 3 Erlangs on 5 servers, Erlang B says 11.0% are turned away; Erlang C says 23.6% wait 43 sec on average.
The same load, two assumptions
30 customers arrive every hour at random and each occupies a server for 6 minutes on average, so the load is a = 30 × 6 ÷ 60 = 3 Erlangs: 3 servers busy on average. There are 5 servers. What happens to a customer who finds all 5 busy is the only difference.
Blocking probability = 11.01%. The engine, with a waiting room of 0, turned away 11.02% of arrivals. Nobody waits.
Probability of waiting = 23.62%, average wait 43 sec. The engine measured 23.59% and 43 sec. Nobody is lost.
The formulas
Erlang B: B(c, a) = (aᶜ ÷ c!) ÷ Σₖ₌₀..c aᵏ ÷ k! Erlang C: C(c, a) = B ÷ (1 − ρ × (1 − B)), ρ = a ÷ c (needs ρ < 1) Wait: Wq = C ÷ (c × μ − λ)Erlang C is computed from Erlang B, which is why the two always move together: more servers lower both. Erlang C is always higher than Erlang B for the same a and c, because a customer who waits is counted even if they would have been served shortly after.
How both change with the number of servers
| Servers | Erlang B: turned away | Erlang C: chance of waiting | Erlang C: average wait |
|---|---|---|---|
| 4 | 20.61% | 50.9% | 3.1 min |
| 5 | 11.01% | 23.6% | 43 sec |
| 6 | 5.22% | 9.9% | 12 sec |
| 7 | 2.19% | 3.8% | 3 sec |
| 8 | 0.81% | 1.3% | 1 sec |
Load 3 Erlangs (30 an hour × 6 minutes). Simulation: seed 7, 30 replications of 200 hours, exponential service.
Choosing
- Customers who cannot wait are lost: a full parking lot, a fully booked line, a call that gets a busy signal. Use Erlang B, and set the share you can afford to lose.
- Customers wait however long it takes: a queue in a yard, a hold queue with no limit. Use Erlang C, and set a wait or a service level.
- Customers wait up to a limit: a yard that holds 4 trucks, a call queue capped at 10. Use the limited waiting room calculator; Erlang B is its zero-space case.
- Customers wait, then give up: that needs a model with abandonment (Erlang A), which is not included here or in the Simflowly engine.
Both formulas assume random arrivals and identical servers. Erlang B is also exact for any service-time shape with the same mean; Erlang C assumes exponential service, and more regular service gives shorter waits (see Erlang C vs simulation).
Questions
What is the difference between Erlang B and Erlang C?
Erlang B assumes a customer who finds every server busy leaves and never returns, so it gives the probability of being blocked. Erlang C assumes that customer waits in an unlimited line, so it gives the probability of waiting and the average wait. Both need random arrivals; Erlang C also assumes exponential service times.
Which one is right for a call centre or a dock?
Neither automatically; ask what happens when everything is busy. Callers who hang up or trucks turned away from a full yard behave like Erlang B. Callers who hold and trucks that queue in the yard behave like Erlang C. A limited holding room sits between them and is covered by the M/M/c/K calculator.
Do Erlang B and Erlang C give the same number of servers?
No, because they answer different questions. With the same load, 8 servers keep turn-aways at or below 1%, while 6 servers keep the chance of waiting at or below 20%. Compare targets that mean the same thing for your business before you compare the formulas.