Quality of Service Network Classification
Introduction
The
rapid increase in usage of bandwidth hungry peer-to-peer
applications has prompted many network administrators at
large organizations to provide quality-of-service (QoS) at
a local level (with respect to a shared WAN link)
via commercial bandwidth shaping devices such as Packeteer's
PacketShaper, Sitara
Networks QoSWorks, and Cisco's
Network-based application recognition.
These devices target the type of network topology shown at left. Notice that
they do not attempt to provide end-to-end quality-of-service. Rather,
they are concerned simply with providing service guarantees of the shared WAN
link.
These devices rely on level-7 traffic analysis to attribute
traffic to
specific applications. In other words, they look at raw packet
data and
perform pattern matching against a rule list of applications
defined by each
vendor. After identification, a flow is controlled using
low level TCP flow
control mechanisms. The QoS device can use low-level TCP
flow control
mechanisms to
influence transmission rates. |
 |
My Work
My work has involved evaluating the long term viability of this method of identification
as a means of differentiating services. My investigation produced the confounding
socket, a countermeasure to deterministic flow identification that renders
traffic invisible to real-time flow classifiers currently used in commercial
systems. This technique represents a significant weakness in current systems
that will inhibit their ability to provide quality-of-service guarantees in the
long-term.
I presented this work in the ACM Student Research Contest undergraduate division
and was awarded first place at the ACM awards banquet in San Diego, CA. For
more information about the project, check out the links below. My talk gives
a brief overview of my work. The poster provides slightly more depth. For a
thorough treatment, including implications and applications, check out the
paper.
|
|
