Recent research in the networking literature has identified the presence of long-range dependence, and indeed self-similarity, in LAN, WAN, WWW, and video traffic. Long-range dependence implies significant correlations over long time scales, and self-similarity implies similar-looking traffic bursts across many time scales.
Such traffic characteristics pose an interesting challenge to the designers of ATM networks, because it clearly violates the memoryless property assumed in most analytical performance studies. It also begs the question: What happens when you mix together multiple traffic streams, each of which is self-similar?
Our research efforts to date, led primarily by Carey Williamson, have focussed on modeling self-similar network traffic, and then using these simulation models to address the question above. We have developed several theoretical results for the aggregation of self-similar traffic streams, and then confirmed them using simulation, with ATM-TN. We have also used ATM-TN to assess the accuracy of a formula, proposed by Ilkka Norros, for the effective bandwidth required by self-similar traffic streams, and to assess its sensitivity to traffic characteristics.
 Research
Areas |
 DISCUS
|
 Research
Groups |
 Comp
Sci. Dept |
 University |