Soutenance de thèse -- Ítalo Cunha

Bonjour,

j'ai le plaisir de vous inviter à la soutenance de thèse de Ítalo Cunha
intitulée "Tracking Internet Routes and Path Reachability for Network
Tomography".

==============
Date et lieu
==============
Jeudi 7 juillet 2011 à 16h00

4, place Jussieu, Paris
Salle 101, 1er étage, couloir 25-26

==============
Jury
==============
- Mark Crovella, Boston University, USA. (Rapporteur)
- Vern Paxson, UC Berkeley/ICSI, USA. (Rapporteur)
- Krishna Gummadi, Max Planck Institute, Germany. (Examinateur)
- Matthieu Latapy, UPMC, France. (Examinateur)
- Renata Teixeira, UPMC, France. (Directrice de Thèse)
- Christophe Diot, Technicolor, France. (Co-directeur de Thèse)

==============
Résumé
==============
Network failures adversely impact Internet's end-users and service
providers. It is important for end-users and service providers to
diagnose (i.e., detect, identify, and troubleshoot) network failures
quickly and reliably to fix or bypass them. Network operators can use
control data collected at network routers to detect and identify
failures in the autonomous systems that they manage. Unfortunately,
other than in this specific case, control data is not available and is
insufficient to identify all types of failures. Hence, end-users and
network operators resort to active end-to-end measurements to diagnose
failures.

In this thesis, we deal with the first two steps of network failure
diagnosis, namely failure detection and identification. Binary
tomography algorithms hold great promise for helping network operators
detect and identify network failures, including those that cannot be
detected with router-based information. Despite much previous attention
to binary tomography algorithms, two significant obstacles have
prevented network tomography from being applied in practice: (1) the
lack of resources to measure paths frequently enough to detect all path
changes and maintain an up-to-date network topology; and (2) the
inability to distinguish failures from congestion-related losses and the
impossibility to synchronize end-to-end measurements required to build
reachability matrices.

This thesis designs measurement techniques to track path changes and
end-to-end reachability. We show that Internet paths are mostly stable
but sometimes go through instability periods. We exploit this behavior
to design a predictor of path changes. DTrack, our path tracking system,
uses our predictor to focus probing on the unstable paths, and
consequently increase the number of changes detected and shorten the
detection time. We also implement a failure confirmation method to
differentiate packet losses from persistent failures in Internet paths.
We also design aggregation strategies to combine path status
measurements from multiple monitors into a consistent reachability
matrix. DTrack addresses the first obstacle to deploy tomography in
practice, while confirmation and aggregation address the second. We
design and evaluate our techniques using analytical modeling,
trace-driven simulations, controlled experiments in Emulab, and
wide-area deployments in PlanetLab. We show that our techniques allow
accurate identification of all failures longer than a chosen target
duration and reduce the number of false alarms by two orders of
magnitude.

Cordialement,
Ítalo Cunha