A Call Acceptance Algorithm For Gigabit Networks

This paper presents a real time front-end admission control scheme for gigabit networks which dynamically shapes the traffic. In this model the calls, characterized by their mean and variance a priori, arrive with a Poisson distribution and have an exponential duration. The state of the transmission link is completely described by the net mean and variance which increases linearly with the number of calls. It can be reasonably assumed that the traffic on the link tends towards a Gaussian distribution as the number of users increase. Figure l shows the effect of the burstiness on the distribution of the statistically multiplexed traffic on the link. The distribution tends to shift and widen as traffic on the link grows. The traffic is correlated and is akin to white noise passed through a low pass filter. When the demand on the link exceeds the capacity of the link, cells (53 Byte cell) are lost and we can compute the Cell Loss Probability (CLP) analytically by using the Guassian approximation. Figure 2 shows the CLP as a function of the number of users on a 1 Gbps link. Three classes of calls are depicted: Class A relates to the set of calls with a 10 Mbps peak rate at 10% duty cycle (DC), Class B represents non-bursty calls with a 4 Mbps peak with 25% DC, and Class C signifies very bursty traffic with a peak of 25 Mbps at 4% DC (notice that the average data rate is 1 Mbps in all three classes).