EECS 290Q: Project Abstracts

Self - Similar Traffic Models for Communication Networks
Socrates Vamvakos

In a recent series of papers (e.g. [1]), traffic measurements on Local Area Networks (LANs), which are both very accurate and extensive in time, have been reported. These measurements challenge traditional data traffic modelling, which have been mostly based on the Poisson process, or more generally, short - range dependent processes, since one of their most striking features is the tremendous burstiness of the traffic at, practically, any time scale. Moreover, statistical analysis has shown that the traffic is self - similar in nature with a surprising accuracy. As a result, it becomes necessary to study the performance and the design of network components, and in particular storage and switching subsystems, by taking the long - range dependence of the input traffic into consideration.

We propose to read the listed references and try to give a comprehensive review of this relatively new area. We will show some of the implications that these new traffic models have on ATM buffer analysis.

UCB Networking / Matt Siler siler@eecs.berkeley.edu/ April 19, 1996