University of Houston
Department of Computer Science
In partial fulfillment of the Requirements for the Degree of
Doctor of Philosophy
Tsung-i (Mark) Huang
will present his preliminary defense
Fast TCP Throughput Prediction through Time-Series Analysis
Abstract
Predicting long-term bandwidth available for HTTP and other TCP
based transfers from a server on the internet is important for
grid network resource service and management of web content
delivery engines like web caches, content delivery services, and
mirror servers. However, fast and accurate TCP bandwidth
prediction is a challenge. TCP throughput is typically measured
by sending a file between two hosts and recording the transmission
time, but this file should be moderately large before the
bandwidth achieved reflects the long-term bandwidth, and that
implies significant delay and cost. This proposal presents a novel
approach on fast TCP throughput prediction. The key idea is to
use a time-series of wall-clock and data size collected from
one-direction TCP trace at receiver side, apply a partitioning
algorithm to generate a new per RTT time-series, and then, with
knowledge of TCP, long-term throughput prediction of this TCP
stream can be made without waiting till the end of file
transmission. We present a few partitioning algorithms and
preliminary results. And some challenges of this research problem
and work plan are also presented.
Date: Friday, September 5, 2003
Time: 11:00 AM
Place: 550-PGH
Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Jaspal Subhlok