Traffic engineering in IP networks

This work was aimed at better understanding what is achievable using different traffic engineering techniques, and more important how one can devise solutions that are not highly sensitive to the quality of the “input” on which traffic engineering decisions are based. One problem that was investigated included how to best group traffic flows to minimize the number of distinct paths that had to be established in order to achieve optimal performance, and how this grouping affected both long term and short term performance. Another problem focused on devising minor modification to existing IP forwarding in order to allow the implementation of near optimal traffic distribution over existing IP networks. Yet another area of investigation involved evaluating traffic engineering solutions that can provide resilience to changes caused by link or node failures or by fluctuations in traffic patterns. The work was partly carried out in collaboration with Christophe Diot (now at Google) and his former group at Sprint ATL. Another aspect of this work, investigated the use of overlay networks to deliver better service “guarantees” over IP networks by leveraging the (path) diversity offered by the availability of a large number of peers. Part of this investigation was carried in the context of real-time applications such as VoIP and video (see above), but another area of focus was to develop simple techniques for quickly identifying peers that could act as “good” relay nodes for alternate overlay paths, i.e., overlay paths with performance that was largely uncorrelated with that of the default path. The challenge was to retain the benefits of having access to a very large number of possible choices, while devising simple and scalable solutions, i.e., that required little processing and only minimal storage.

Penn Contributors

Publications