Home :: Short Bio :: Contact :: Publications :: Talks :: Tutorials :: Software :: Supervision
K. Pentikousis, ECN, power consumption, and error modeling in TCP simulation studies, PhD Dissertation, State University of New York at Stony Brook, December 2004.
Abstract — Recent developments in wireless access technologies are pointing towards hybrid networks with wired and wireless components integrated in terms of both services and physical architecture. This thesis focuses on three distinct topics, which interrelate to the extent that each has an important bearing on the research and investigation of the interplay between this emerging hybrid wired/wireless Internet environment and the Transmission Control Protocol (TCP).
First, we study the effect of explicit congestion notification (ECN), focusing on situations in which all nodes in the network support ECN. The results show that ECN does not necessarily lead to significant improvement in TCP goodput. On the other hand, we show that network resources are used more efficiently and more fairly. In particular, TCP flows benefit from lower overhead for unsuccessful transmissions and lockouts are largely avoided. Finally, we show that global synchronization is no longer an issue, and argue that current TCP versions have essentially solved the problem.
In the emerging wired-cum-wireless environment, battery-powered mobile hosts play a central role and power consumption becomes a critical metric in evaluating network protocols. We develop tools to measure protocol energy efficiency in simulation studies, which take into consideration the battery recovery effects and communication patterns, including the amount of data transmitted and received and the time spent idling.
The third and final focus of our work is the development of a TCP-centered error model based on real network traffic traces. We present a methodology for calibrating a temporal error model for TCP simulation studies based on the analysis of IP packets monitored at the egress point of stub autonomous systems. Trace data is distilled into alternating ``good'' periods, and ``bad'' periods during which in-flight data segments are lost. Heuristics are developed for dealing with periods of no activity. Results are presented in which the lengths of ``good'' periods are fitted to a Pearson Type VI distribution, and the number of serial occurrences of such periods is fitted to a ``discretized'' Lognormal. The lengths of ``bad'' periods are shown to possess a non-standard distributional form.
Full Text—UMI (ISBN: 0-496-94465-7)
Tags—Transport protocols, Traffic measurements and analysis, Simulation and modeling
Home :: Short Bio :: Contact :: Publications :: Talks :: Tutorials :: Software :: Supervision
Last updated: 22.11.2009