Conclusion

IP networks are under increasing pressure to provide predictable communication performance for applications such as voice over IP, interactive gaming, and commercial transactions. These applications are sensitive to both transient disruptions (i.e., during routing changes) and persistent congestion (i.e., when the routing does not match the prevailing traffic). In this paper, we propose a new mechanism for selecting egress points that satisfies both requirements. TIE avoids the disruptions caused by hot-potato routing changes while supporting diverse network-wide objectives such as traffic engineering and maintenance planning.

TIE is simple enough for routers to adapt in real time to network events, and yet is much more amenable to optimization than today's routing protocols. In addition, TIE can be deployed in an AS without changing the intradomain or interdomain routing protocols, and without the cooperation of other domains. Our experiments for two network-management problems, using data from two backbone networks, demonstrate the effectiveness of our new mechanism and the ease of applying conventional optimization techniques to determine the best settings for the tunable parameters.

In Sections IV and V, we evaluate TIE on data from two operational networks. In this appendix, we present our methodology for obtaining the input data--the internal topology, the egress sets, and the traffic demands--from passive measurements. Since routers in the same Point-of-Presence (PoP) essentially act as one larger node, we model the topology of both networks at the PoP level.