Simulation Study of Proposed HighSpeed TCP for Large Congestion Windows
|
|
Simulation Study of Proposed HighSpeed TCP for Large Congestion Windows |
|
|
Author - Evandro de Souza In this project I have been studying the behavior of Sally Floyd's
proposal for HighSpeed TCP in high speed long distance links. The HighSpeed TCP - HSTCP is an
on going work that proposes a modification to TCP's congestion control
mechanism for use with TCP connections with large congestion windows.
HighSpeed TCP is expected to provide an effective mechanism for
bulk data transfer, while maintaining fairness with other types of TCP
already in use. The development of this work is part of the requirement
to obtain my master degree at State University of Campinas -
UNICAMP , Brazil, in the Department of Computer
Engineering and Industrial Automation - DCA , under advising of Eleri Cardozo . This
work has been conducted at the Lawrence Berkeley National Laboratory
- LBNL , USA, in the Distributed
System Departament - DSD
. |
|
. The simulations of HighSpeed TCP have all been carried out using the NS-2 Simulator , which has been modified to include the HighSpeed TCP. The NS-2 version used is ns-2.1b9 released on April 22, 2002 and more recent updates available at NS-2 repository . The network topology used is a simple single bottleneck network, also know as "dumbbell". This topology is very simple to understand and to see the effects of the interaction between the flows. The bottleneck link bandwidth used in the simulations is 1 Gbits/s and a link delay of 50 ms. Simulations using both RED and DropTail router queue management were performed. |
|
|
We used three sets of primary flows in most of this study. The first set had only HSTCP flows, the second was composed only of REGTCP (standard TCP) flows, and the third set contained a combination of REGTCP and HSTCP flows. The number of flows for each set varied according to the experiment. The three sets of flows were each exposed to different network environments. In the first network environment there were no other traffic sources and no extra interference beyond that generated by the REGTCP and HSTCP flows. This network environment we refer to as Ideal Condition. The second network environment represented the situation where there were systemic losses (or losses not directly related to congestion). We call it Lossy Link Condition. Some number of packets were randomly dropped from the flows, with a defined drop rate. The third network environment explored the reaction of the flow sets to bursty traffic, so we refer to it as Bursty Traffic Condition. The bursty traffic was composed of short-lived standard TCP flows running for a few seconds. Each simulation was run for 300 seconds with a variable number of HSTCP and REGTCP flows, depending of the objective of each experiemnt. The reports from these simulations can be seen here . For more information about this work, please contact Evandro de Souza or Deb Agarwal . |
|
Page last modified:
Monday, 30-Jun-2003 11:18:51 PDTContact: Webmaster < webmaster@george.lbl.gov
> Credits:The research and development of the Distributed
Systems Department is funded by the U.S. Dept. of Energy, Office of
Science, Office of Advanced Scientific Computing Research, Mathematical,
Information, and Computational Sciences Division. Privacy and site security
notice to Users You are visitor [an error occurred while processing this directive] to this page since June 2002. |