I've heard (or may be read somewhere) that Erlang loss model (M/M/n/n queue) is fully insensitive to distribution of holding (sojourn/stay) time. Thus, we can use the famous formula of M/M/n/n loss system (including blocking probability and state probability) for M/G/n/n loss system.

P(k)=P(0) x (lambda/mu)^k / k!

P(k) = Probability of k customer in M/M/n/n queue

P(0) = Probability of queue being empty

lambda = arrival rate

mu = service rate

(Above from page 105 of 

Kleinrock, L. (1975). Queueing System. Wiley-Interscience. See attachment)

I would be grateful if anybody knows a reference introduce it to me. Please mention the the exact page number.