4 ms·

M/M/s queues have nice properties, but require that arrival and service times are exponentially distributed. Once you deviate from this assumption, closed-form

M/M/s queues have nice properties, but require that arrival and service times are exponentially distributed. Once you deviate from this assumption, closed-form solutions become difficult to obtain and numerical approximations are quite messy. I've therefore always wondered whether there is software that uses theoretical results from queueing theory to improve control or to optimize parameters of complex systems. Any references?

kqr 4y ago
...except! If the server is processor-sharing/timeslicing. The results derived for timeslicing servers are surprisingly general in service distribution, and can be stated in product form like the networks of simple M/M/c servers you know. (Arrivals still need to be Poisson, but that's a fairly nice requirement in that many real-life arrivals actually look Poisson.)