|
Nina Taft
· I joined Intel Research in Berkeley in October 2003 and my current research agenda is primarily focused on network security, distributed monitoring, and traffic characterization, with particular interest in enterprise networks. Prior to joining Intel, I spent nearly five years at Sprint, in the IP Group at the Advanced Technology Laboratories in Burlingame, CA. At Sprint I conducted research on numerous traffic engineering problems such as traffic matrix estimation, Tier-1 traffic characterization, capacity planning and routing. Prior to Sprint, I worked at SRI International in Menlo Park, CA for four years. Most of my work then focused on congestion control and QoS-routing in ATM networks. I received my PhD from the University of California at Berkeley in 1994, under the supervision of Prof. Pravin Varaiya. My thesis title was: "Performance Analysis of Broadband Networks: Congestion Control and Traffic Characterization." One of the more fun bits of my thesis involved pursuing the benefits of using entropy as a traffic descriptor for ATM networks.
My research agenda focuses primarily on the following. (1) I am interested in the network security problems that large enterprise networks face. We use data mining techniques for networking security solutions that build on statistical analysis of traffic patterns. We're interested in frameworks that incorporate data mining solutions into distributed monitoring systems. Some projects focus on detecting network-wide attacks by tracking features that represent a global state of the network. Other efforts focus on improving a laptop's ability to protect itself when it leaves the heavily infrastructure protected environment of an enterprise. (2) I am still working on traffic matrices although less on estimation and more on how to synthetically generate them. Synthetic traffic matrices are needed for performance evaluation studies since they are the input to many algorithms our community designs. (3) I am also interested in the interaction of overlay networks with each other and with underlying carrier networks. We study the race conditions that can occur when multiple overlays co-exist simultaneously and the goal is to understand whether or not such race conditions are cause for concern or mandate a change in the way we design overlays. Want to know how to generate a synthetic traffic matrix for your performance evaluation study? Traffic generation at a single link has its uses, but it doesn't give you the ability to load up an entire network with traffic - as a traffic demand matrix can. See our CCR paper for details [PDF]. This paper outlines the extent of the problem and its nuances, plus presents one solution. Want an overview of the traffic matrix literature? See: "The Evolution of Traffic Matrix Techniques and Applications: Past, Present and Future". [ppt] This talk was initially given as the keynote at the Large Scale Network Inference (LSNI) workshop co-located with ACM Sigmetrics in June 2005. This talk has been presented a few times in various seminar series, including ICSI, HP Labs, Intel Folsom, UC Riverside and the University of Melbourne.
See below for a listing by subject: | ||||||||
|
Executive Committees:
Program Committees:
| ||||||||
