Raj's Ongoing Projects
Fault Detection and Identification in Routing Protocols
In any setting that involves a network, many important computations
are carried out in a distributed fashion. Examples are
routing-algorithms, parameter-estimation algorithms and distributed
computations. These distributed algorithms assume cooperating,
error-free behavior among the nodes; i.e. each node involved in the
computation carries out its tasks faithfully. This assumption is
valid when the nodes are homogeneous, are present in a benign
environment, and under central control. However if the environment is
hostile allowing nodes to be compromised, or if nodes are
heterogeneous allowing for different interpretation and implementation
of the algorithm or if there is no central control that can verify and
ensure the validity of each node's execution of the distributed
algorithm, the computation may fail. Whether a distributed computation
will fail, the conditions under which it will fail, and the extent to
which it will fail are important questions, answers to which will help
us determine the robustness of the distributed computation. Until now
research and analysis of distributed algorithms has focused on the
convergence properties of distributed algorithms, while assuming that
nodes in the computation function as they should. In the real world,
nodes malfunction or can be compromised, especially in the emerging
areas of sensor-nets and ad-hoc networks. It is important and useful
to analyze the properties of distributed algorithms under such adverse
conditions. In my work I am studying such resilience analysis.
I first deal with the issue of defining, classifying and if possible,
quantifying the notion of resilience in a distributed algorithm.
Then I focus on a particular aspect of resilience I call detection and
grapples with the issue of whether and when nodes can sense that there
is an error in the distributed computation. Toward such detection, we
demonstrate preliminary results which show that, for a distributed
protocol, certain classes of disruptive behaviors are detectable while
other classes of disruptive behavior are not.
Related Papers
-
- Raj Kumar Rajendran, Vishal Misra, Dan Rubenstein,
"Strong Detection of Misconfigurations",
Principles of Distributed Computing, Las Vegas, July 2005.
ps,
pdf,
People
Raj Kumar Rajendran
Vishal Misra
Dan Rubenstein
Overlay Routing For VoIP
Voice traffic on the Internet (VoIP) is poised to grow rapidly
but the Internet is not engineered for such delay-sensitive applications.
In our OverPhone Project we show how the end-user quality of VoIP
calls can be improved by rerouting them in an intermediate overlay
network and thought the use of various path-diversity and error-coding
techniques. We introduce our overlay network (OverPhone) that is tuned
to optimize VoIP quality and provide experimental results of its
implementation on the PlanetLab testbed. We show that re-routing VoIP
calls through OverPhone improves the quality of more than a third of
the paths by significant amounts when using the G.711 codec. We
believe that this benefit extends to other well known and proprietary
codecs as well. We also investigated the benefits and trade-offs of
using techniques such as path-diversity and parity-coding. We found
that path-diversity is useful in sustaining throughput at high loads,
and that parity-codes can be used to trade-off bandwidth for path-quality.
Related Papers
-
- Raj Kumar Rajendran, Samrat Ganguly, Rauf Izmailov, Dan Rubenstein,
"Performance Optimization of VoIP using an Overlay Network",
Columbia University Technical Report .
ps,
pdf,
ppt,
Software
We implemented the OverPhone system in Perl and tested it over 40
nodes on the PlanetLab overlay. We found that the quality of about 30%
of the calls could be improved. The Overphone software is provided
below so that you can test it yourself.
-
Return to main page