A loss (blocking) queuing model—commonly denoted M/M/c/0—describing a system with Poisson arrivals, exponential service times, c identical servers, and no waiting space; arriving customers that find all servers busy are blocked (lost). The model yields the steady‑state blocking probability as a function of offered traffic and c (Erlang B formula).

Principle

Under Poisson arrivals and exponential service, the probability an arrival is blocked depends only on offered traffic (traffic intensity in Erlangs) and the number of servers c; adding servers reduces blocking nonlinearly as given by the Erlang B recursion/formula.

Demonstration

Illustrative scenario → A telephone trunk group accepts calls that cannot queue. Recognition → Offered traffic a (Erlangs) and server count c are identified. Action → Use Erlang B to compute blocking probability and choose c to meet a blocking target. Consequence → If blocking target is met, the proportion of lost arrivals is bounded by the computed probability; otherwise capacity decisions are inadequate.

Misapplication

Applying Erlang B when arrivals are not Poisson, service times are non‑exponential, customers will wait or retry, or when retrials/queueing occur. The error treats the measured blocking as if it arose from a loss system while the system allows patience or retrials, leading to underestimation of required capacity.

Consequence

Correct application gives a tractable capacity rule linking offered traffic to blocking risk; misapplication underestimates carried load or masks congestion modes (queues, retrials) and can lead to insufficient provisioning or incorrect performance forecasts.

Reversal

If the system permits waiting (a queue) or customer retrials, or service times lack the memoryless property, alternative models (e.g., M/M/c with queueing, Erlang C, or models accounting for retrials) are required because the loss‑system assumptions no longer hold.

Boundary

Clearly within: systems with Poisson arrivals, exponential service, c parallel servers and no queue (loss system) analyzed in steady state. Boundary case: short finite buffers where blocking approximates loss behavior only under low utilization. Clearly outside: systems with waiting room, retrials, non‑Poisson arrivals, non‑exponential service, or transient analyses outside steady state.

Semantic Tension

Tractability versus realism: Erlang B's analytic convenience and closed‑form blocking measure trade off against its restrictive assumptions (no queueing, memoryless service, Poisson arrivals); improving realism typically breaks the closed form and requires more complex models or simulation.

Synthesis

Erlang B isolates capacity planning for pure loss systems: it translates offered traffic into a blocking probability under strict stochastic assumptions, so its utility rests on matching those assumptions to operational behavior or recognizing when they fail.