 |
Hung X. Nguyen , PhD student at EPFLLaboratory for Computer
Communications and Applications (LCA) Address: BC 248, Station 14, EPFL CH-1015 Lausanne, Switzerland |
Short Biography Research Publications Teaching
I received the B.Eng. degree in Information Technology and
Telecommunications Engineering from the University of Adelaide,
Australia in 2001 and completed the Pre-Doctoral school in
communication systems at the Swiss Federal Institute of Technology,
Lausanne (EPFL) in June, 2002. I have been a
research assistant since October 2002 and a PhD student since March 2004 at the Laboratory for Computer Communication and
Applications (LCA), EPFL. My advisor is Professor Patrick
Thiran.
I am interested in both the theory and system aspects of computer networking. During my PhD study, I work mainly with Internet tomography and failure diagnosis. One of the weak points of today's Internet is the lack of built-in tools to identify and correct performance anomalies such as routing failures, excessive delays or losses. Calculating link performances at a particular time using end-to-end measurements is called network performance diagnosis ; whereas inferring parameters of the statistical models governing these performances is called network performance tomography. Both network diagnosis and tomography are notoriously difficult because regular unicast data do not provide enough information to allow the identification of the link performances. More specifically, the system of equations relating end-to-end performances with link performances is under-determined and hence has infinitely many solutions.
I study three fundamental questions of Internet tomography and diagnosis using
regular unicast probes. First, what is the optimal end-to-end probing strategy for network diagnosis and tomography?
Second, how do we overcome the under-determined problem of network diagnosis
to accurately troubleshoot performance issues? Third, which statistical
properties of network links can be inferred from end-to-end measurements?
Our main methods exploit the spatial and temporal
distribution of packet losses
on the Internet. Namely, most links on the Internet have negligible
loss rates and that links with high loss rates usually also suffer high loss rate
variances. Details on theoretical findings and their implementations can be found under the
following link.
I also work on randomized algorithms for overlay routing. We analyze the performance of the famous two random choices method on load balancing overlay networks. We show that the simple two random choices algorithm does not work on many realistic settings and propose modifications to this algorithm that provides guaranteed performance on practical overlays. Latest results can be found under the following link.