[home] [research] [courses] [publication] [contact]
I design, analyze and evaluate traffic engineering techniques in large IP backbone networks, with a focus on efficient and robust traffic engineering in a dynamic network environment.
REIN is a solution framework that can be incrementally deployed in the Internet and improve the reliability of IP networks at low cost. The key observation is that potential redundancy exist across different networks. REIN allows neighboring networks to use the resources of each other as backup, and has the potential to improve network reliability at low cost. Using real traffic traces and network topologies, we demonstrated that a network can substantially improve its reliability by using only a small amount of backup resources from its neighbors.
For many networks, during common scenarios, traffic is relatively stable, but during some rare, unexpected scenarios, traffic can be highly dynamic. Traditional traffic engineering approaches either optimize for the common scenarios or the (worst-case) unexpected scenarios, but not both. COPE is a class of intra-domain traffic engineering algorithms that optimize performance for common scenarios and provide worst-case guarantee for unexpected scenarios. Evaluations using real Internet traffic traces and topologies demonstrate that COPE out-performs state-of-the-art intra-domain traffic engineering algorithms.
We studied the impact of inter-domain traffic engineering on the stability of BGP for the current Internet, and proposed guidelines for inter-domain traffic engineering to guarantee routing stability without global cooperation among different networks.
Multi-homing is a popular method used by large enterprises, stub ISPs, and even small businesses to connect to the Internet in order to reduce cost and increase performance. Given a set of locally available ISPs, a user may want to choose the best set of ISPs to minimize the cost. This is referred to as the optimal ISP subscription problem. We designed a dynamic-programming algorithm to solve this optimization problem. Using real traffic traces and realistic pricing model, we demonstrated that our algorithm solves the ISP subscription problem optimally.