Performance Analysis of a Statistical Multiplexer Offered Integrated Voice and Data Traffic

In order to adequately traffic engineer resources in an ISDN environment, it is important to have predictive models which relate system performance to the offered traffic characteristics. For example when engineering statistical multiplexers, offered a superposition of packetized voice processes and data traffic, traffic variability and correlations need to be taken into account because of the significant effect these factors have on system performance. The effects are particularly strong on the tail of the voice packet delay distribution, which usually has a stringent requirement. Predictive models are of particular interest for obtaining these low probability tails which can be difficult to obtain from simulations. In this paper we study the traffic characteristics and queueing behavior of multiplexers for packetized voice and data. The input process is a fairly complex process and possesses correlations that result from the fact that the aggregate voice packet arrival rate is a modulated process obtained by modulating the individual voice source packet rate by the number of voice sources in their talkspurt. Even if a component voice process is modeled as a renewal process, with a geometrically distributed number of deterministically spaced packets during a talkspurts followed by an exponentially distributed silent period, the superposition process is a complex, non-renewal process and exact analysis is intractable.