ABSTRACT
The intensity of data traffic arriving at a point can be specified in various ways such as bits per second, packets per unit time, ordered pair of ON time for a packet and OFF time between successive packets, or, the total number of bits or packets received over a given large time interval. Unless otherwise specified, there is the implicit notion that these intensity figures do not change from one time interval to the next, and that the quantities in successive intervals are independent, over meaningful ranges of times. The observed traffic in some types of data networks appears to violate this property. Packets may arrive in clusters, with occasional packets present between clusters separated by appreciable time intervals. These occasional packets may not allow us to model the incoming traffic with intermittent time periods of arrivals. Furthermore, the variance in the amount of traffic may not decrease rapidly as the length of the time interval used for averaging. Therefore, averaging the arrivals over a long time interval to include several packet clusters in the interval does not help us in realizing stable values for intensities. Traffic exhibiting such characteristics is referred to as bursty traffic. The two causes of traffic burstiness are (1) unbounded variance, and (2) long range dependence (LRD) in interarrival times, described below.
