Good News about Heavy Tails
Joachim Charzinski
Abstract
Heavy-tailed distributions have been widely found in traffic characteristics
like connection volumes or connection durations in today's data networks.
The purpose of this paper is to show that these heavy-tailed distributions,
besides the well-known undesirable features like producing self-similar,
long-range dependent traffic, also exhibit desirable features in some cases.
Two such cases are presented here: (1)~the detection of flows in network
traffic, making it possible to determine the right point in time
to establish a shortcut connection for a flow. (2)~the dimensioning of
dial-in access servers where the blocking probability is actually reduced
when the holding time variance increases.
Keywords
Traffic Engineering, Network Dimensioning, Traffic Characteristics,
Real User Networks,
Heavy Tail, High Variance, Residual Lifetime, Access Server Dimensioning