Title page for ETD etd-05202004-152121

Type of Document

Dissertation

Author

Oothongsap, Phoemphun ,

Author's Email Address

poothon@ncsu.edu

URN

etd-05202004-152121

Title

Analysis of High-Speed Data Transfer Protocol Algorithms

Degree

PhD

Graduate Program

Computer Engineering

Advisory Committee

Advisor Name	Title
Ioannis Viniotis	Committee Chair
Mladen Vouk	Committee Co-Chair

Keywords

bandwidth
MIMD
high speed long delay network
TCP/IP
congestion control

Date of Defense

2004-04-27

Availability

unrestricted

Abstract

High-performance networks offer the promise
of connectivity at speeds of 40 Gbps or more. However, effective use of
the bandwidth they offer is a challenge since no reliable
transport protocol is well suited for such environments.
The most popular reliable protocol currently available is standard TCP/IP.
However, studies show that standard TCP performs poorly in high bandwidth
long delay networks, i.e., in long distance backbone networks.
The maximum bandwidth utilization of TCP in such often environments
is less than 10% of the provided bandwidth.
Thus, to utilize the provided
bandwidth effectively, new protocols and algorithms have been proposed.

SABUL is one of the hybrid protocols designed to overcome TCP performance
issues in high bandwidth, long delay environments. SABUL uses a variant of
MIMD (Multiplicative Increase and Multiplicative Decrease) as its
congestion control algorithm, and implements loss rate as a congestion
indicator.
Unfortunately, assessment of SABUL properties is mostly empirical,
and not comprehensive. Similar appears to be the case with many
new protocols being developed in this space.
While emprical work is very useful, it can
very easily miss some key behavioral aspects of the protocol,
especially when it is used in diverse situations.
This work remedies this problem by suggesting
a generalizable framework for systematic study of protocols intented for
use in a high bandwidth, long delay environment. The framework is presented through a
full theoretical, and some experimental, assessment of SABUL congestion control algorithm.

The framework is divided into three sections: mathematical
analysis of the (SABUL) congestion control algorithm, simulation and
experimental analysis of the (SABUL) congestion control algorithm, and
improvement of the (SABUL) congestion control algorithm. The mathematical
analysis is done by using deterministic and stochastic models. These
models are
used to assess (SABUL) properties such as bandwidth utilization,
self-fairness, and aggressiveness/responsiveness. To validate
mathematical models, simulations and experimental analyses are performed.
The results explain SABUL
throughput oscillations, derive bounds on its
aggressiveness/responsiveness,
show that SABUL can be self-fair, and identify conditions under which SABUL
may exhibit excessive packet loss.

Moreover, the results show several
drawbacks of SABUL such as high rate fluctuation and rate synchronization
which cause loss of throughput which was not observed in the analytical model but
was observed in subsequent experiments. Based on this information, we suggest how
SABUL congestion control algorithm can be modified to improve its bandwidth
utilization.

Files

Filename		Size	Approximate Download Time (Hours:Minutes:Seconds)
Filename		Size	28.8 Modem	56K Modem	ISDN (64 Kb)	ISDN (128 Kb)	Higher-speed Access
	etd.pdf	2.74 Mb	00:12:40	00:06:30	00:05:42	00:02:51	00:00:14