TU Berlin

Internet Network ArchitecturesBook Chapters

Page Content

to Navigation

Publications by Type: Book Chapters

Characteristics of TCP Connection Arrivals
Citation key F-CTCPCA-02
Author Feldmann, Anja
Title of Book Self-Similar Network Traffic and Performance Evaluation
Pages 367–299
Year 2002
ISBN 978-0-471-31974-0
Online ISBN 9780471206446
DOI http://dx.doi.org/10.1002/047120644X.ch15
Location Purdue University, West Lafayette, Indiana, USA; AT\&T Labs-Research, Florham Park, New Jersey, USA
Editor Park, Kihong and Willinger, Walter
Publisher John Wiley & Sons, Inc.
Chapter 15
Abstract Packets are the basic unit of the Internet. Yet, most user operations involve more than one packet and user experience depends on the performance of the network on a set of packets. Thus it is not surprising that in today's Internet protocol (IP) networks sets of packets are starting to be used as the basis for network operations. The goals of the network operators include improving traffic engineering by better load balancing and moving beyond best effort services. Operating on sets of packets is more efficient for network elements. In fact, per packet decisions may lead to undesired or even unstable solutions. At the edge of the Internet the transmission control protocol (TCP) offers the abstraction of a TCP connection, each consisting of a set of packets. Within the center of the network, or for non-TCP traffic, a related connection abstraction is provided by an IP flow, a group of related IP packets that are close in time.// In this chapter we demonstrate that the TCP connection arrival process is bursty. We show that the arrival process is asymptotic self-similar. Self-similarity of the TCP connection arrival process implies that the use of standard models in evaluating the performance of resource allocation methods can yield misleading results. Therefore, we characterize TCP connection interarrival times using heavy-tailed distributions. We present statistical evidence that such distributions, especially the Weibull distributions, yield a better model for the interarrival times of TCP connections than exponential models. Intuitively, a heavy-tailed interarrival time means that if no connection arrived for some time it becomes more and more unlikely that one will arrive soon. This holds even if the underlying arrival process is nonstationary. Finally, based on a simple resource allocation problem, we show that there are advantages to using Weibull distributions to model TCP connection interarrival times over a nonstationary Poisson process.
Download Bibtex entry


Quick Access

Schnellnavigation zur Seite über Nummerneingabe