A study of the effects of bursty traffic

We study the effects of bursty Internet Traffic through simulations. Both short-range dependency (SRD) and long-range dependency (LRD) traffic with self-similar behavior are simulated over different burst parameters. The results are collected for 10 different 24 hour periods in order to study day-to-day statistical fluctuation. The performance fluctuation from day to day is measured. Effects of employing different traffic admission constraints are examined. An alternative to improve network throughput and utilization is proposed. Finally, a case when arrival patterns of traffic are correlated is explored. Our results show that when traffic behavior and service rate are probabilistic a network has very little control of the situation and the best thing to do is to drop packets when a network experiences too many "late" packets. However, different restrictions used to manage traffic admission yield different trade-offs.