Performance analysis of the general renewal multi-server finite waiting room model

In this paper, we analyze the performance of GI/GI/s/s+r systems. In particular we examine steady-state call blocking and time blocking. We vary several parameters such as the number of servers, the number of waiting spaces, the squared coefficients of variation of the arrival and the service processes, and the third moment of the arrival process. We show that these parameters significantly affect system performance. Thus, for most systems with bursty and skewed traffic patterns, simpler models are insufficient. However, in certain circumstances, when the arrivals are either highly bursty or highly skewed, the Markovian assumption gives acceptable approximations to the GI/GI/s/s+r model. Finally, we observe a behavior that is similar to Erlang's insensitivity property in the M/G/s/s model.< >