UCSD Jacobs School of Engineering University of California San Diego
Pulse Subscribe | Archive | Survey
 

Handling a Hot Potato in Network Routing

Renata Teixeira (Ph.D., computer science, ‘05) at AT&T Labs where her analysis of network disruptions is changing the way data is routed over the Net.

Renata Teixeira worked in AT&T's New Jersey labs for five months in 2003, and again in 2004. She offers a case study in what can happen when graduate students are given a chance to intern at technology companies. With AT&T colleagues including Aman Shaikh, Jennifer Rexford (now at Princeton) and Tim Griffin (now at Cambridge), Teixeira influenced the Fortune 500 company and an entire industry. "AT&T gave her access to all the routing data from its domestic backbone to analyze the network's behavior," says computer science and engineering professor Geoff Voelker, her faculty advisor. "Looking at the data, Renata developed unique insights that are already changing the way the telecom giant, router manufacturers and large Internet service providers handle the way they route data over the Net."

What Teixeira found was that there is a fundamental disconnect in today's routing protocols when they have to cope with so-called ‘hot-potato' disruptions. She observed that an event happening inside one domain can trigger an enormous number of routing changes on traffic flowing between domains, which could lead to packet losses and longer delays. "The interaction of these two protocols and its impact on end-to-end performance was not very well understood," explains Teixeira. "It might only take tens of milliseconds to route around disruptions in AT&T's own network, but the response might take as much as a few minutes when it triggers inter-domain changes. You can imagine how serious a problem that is for anyone using voice over IP, or someone watching streaming video."

At SIGCOMM 2004, she delivered a paper outlining an analytical model of hot potato routing that incorporates metrics to evaluate network sensitivity to hot-potato disruptions. "There may be road construction somewhere and the fiber gets cut and the link has to go down," explains Teixeira. "With the model we built, network operators can predict exactly what the effects are going to be and try to use the model to minimize disruption to end users."

The research at AT&T led to a doctoral dissertation that the Brazilian-born Teixeira successfully defended in February. AT&T has now asked vendor companies to reduce convergence times on future routers, and Teixeira could have fresh innovations up her sleeve: she's developing a tool that operators could use to improve the robustness of their networks.