T.S. Eugene Ng
Carnegie Mellon University
About the talk:
To optimize the performance of large-scale distributed applications such as Napster or Gnutella file sharing and end system multicast, one must know the performance of a large number of underlying network paths. Although active network measurement can accurately determine network performance, it suffers from scalability and timeliness limitations. We believe the ability to accurately predict network distance (round-trip transmission and propagation delay) without active measurement is a fundamental building block in enabling scalable, fast, and effective performance optimization in such large-scale distributed applications.
In this talk, I will present a new approach to predict Internet network distance called Global Network Positioning (GNP). The key idea is to model the Internet as a geometric space (e.g. a 3-dimensional Euclidean space) and distributedly compute geometric coordinates to characterize the positions of hosts in the Internet. The modeled geometric distances between hosts are then used to predict the network distances. I will contrast GNP to the state-of-the-art network distance prediction approach called IDMaps and argue that GNP is more scalable and more nimble. I will also discuss a variety of technical issues in GNP such as what geometric space model fits the Internet well. Through Internet experiments, I will show that the geometric distances implied by the GNP host coordinates can accurately predict Internet network distances.
About the speaker:
T.S. Eugene Ng is a Ph.D. student in Computer Science at Carnegie Mellon University. His current research focuses on scalable Internet network performance prediction and studying its benefits to a variety of applications.