MINC
Multicast-based Inference of Network-internal
Characteristics
Project Summary
Future high-speed networks will be significantly larger and more complex
than existing networks. The variety and interaction of the applications,
middleware, transport protocols, routing protocols, and router/switch resource
management algorithms will make the design, development, control and management
of the Next Generation Internet (NGI) exceptionally difficult.
A fundamental ingredient in the successful design, control and management
of coming networks will be the accurate measurement and characterization
of its dynamics. This project addresses this problem by proposing new fundamental
research and based on measuring and analyzing the end-to-end performance
of multicast probe traffic in order to infer the performance of individual
links within the network. The key to this paradigm is that multicast traffic
introduces correlation in the end-to-end performance measured by receivers.
This correlation can, in turn, be used to infer the performance of the
links within the multicast routing tree spanning the sender and receivers.
Coupled with a well-designed network measurement infrastructure,
this will permit large-scale analysis of network conditions. Furthermore,
because this analysis yields fine-grained (per link) information, the results
can then be composed to estimate network conditions in parts of the network
not directly instrumented. Our proposed research focusses on the following:
-
Development of statistically rigorous estimation techniques for
known multicast distribution trees. We propose to develop and evaluate
parametric and non-parametric estimators for link-level performance metrics
such as loss rate, delay ststistics, and the available bandwidth of the
bottleneck links when the multicast distribution tree is known. The quality
of thesse estimators will be verified theoretically, through simulation
and empirically on the existing Internet.
-
Development of statistically rigorous techniques for identifying the
topology of an unknown multicast distribution tree. We propose to develop
and evaluate techniques for identifying the complete structure of the multicast
tree when only partial information is available. Our goal is to develop
efficient algorithms for classifying the tree based on end-to-end loss
and delay behavior observed at the end hosts.
-
Development of a multicast inference network tool. Using the above
techniques, we develop a validated, distributed software tool for estimating
link-level performance and identifying bottleneck links. This application-level
module can be used on top of a measurement infrastructure such as NIMI
(the National Interent Measurement Infrastructure) or within a specific
application for the purpose of adapting to network behavior.
-
Development of a network infrastructure to support multicast-based inference.
We develop a network measurement infrastructure that supports multicast-based
network inference. This will be accomplished by extending NIMI to
support multicast-based measurement.
Return to MINC page.
towsley@cs.umass.edu