Relationship between packet loss probability and burst parameters for two types of traffic

A call admission control (CAC) method based on the equality of two types of traffic has been proposed for minimizing the total call-blocking rate. However, application of this CAC method to a packet switched network may increase packet loss because packets have bursty characteristics. The length of time in which bursty arrivals continue is called the talk spurt in ON-OFF modeling of the packet arrival process. When the talk spurt is short, the number of arriving packets per unit time (the "burst density") increases under the assumption in ON-OFF model that the average bandwidths of the flows are fixed. This can cause packet losses. Therefore, the relationship between the packet loss probability and the talk spurt must be analyzed because burst density affects packet loss probability. However, the characteristics of the packet loss probability when the talk spurt changes for two types of traffic need to be clarified. In this paper, we link the CAC parameters with the burst parameters for two types of traffic, narrowband flows and broadband flows, in which the bursty traffic follows the Markov-modulated Poisson process (MMPP). Furthermore, we numerically analyze the characteristics of the packet loss probability for a given number of flows using queueing theory. The results quantitatively show that the packet loss probability is substantially higher when the burst density is higher. They also show that the packet loss probability is drastically affected by the burst density of broadband flows because their bandwidth is wider.