J'ai le plaisir de vous inviter à ma soutenance de thèse intitulée
"Scalable Routing in Provider-Provisioned Virtual Private Networks".
=============
Date et lieu
=============
Mardi 21 Décembre à 11 heures dans la salle 101, couloir 25/26, 1er étage.
Vous êtes également invités au pot qui suivra.
=========
Abstract
=========
Companies often have sites spread in distant locations that need to
communicate. Instead of building their own communication network, most
companies today prefer to contract a Virtual Private Network (VPN)
service. In this service model, a VPN service provider shares its
network infrastructure among multiple enterprises in order to
interconnect their distant sites, reducing thereby the communication
costs for each enterprise. Routing is the main task of VPN service
providers; it ensures the communication between VPN sites. The
performances and the robustness of routing are crucial for both VPN
customers and providers. VPN customers need availability, whereas VPN
providers, on the other hand, want a scalable solution to connect the
maximum number of customers. The most common provider-provisioned VPN
technology uses the Internet inter-domain routing protocol BGP for
routing. In Internet backbones, BGP slow convergence can hurt
availability and BGP routing table sizes have grown exponentially over
time, which threatens the scalability of the Internet whole routing
system.
Although they use the same routing protocol, VPN backbones are very
different from Internet backbones. This thesis studies how this
difference impacts the routing convergence and scalability of VPN
provider backbones. First, our measurements on a large VPN provider
backbone reported a route explosion phenomenon similar to that
observed on Internet backbones, if not worse. Some routers in such
networks have to handle from three to almost ten times the number of
BGP routes that a core Internet router handles. This route explosion
has a direct impact on the resiliency of such networks. In fact, some
failures can cause routers to loose a large fraction of their routing
tables. Our measurements showed that, due to the high number of
routes, routers can spend an immoderate amount of time to receive the
missing routes and recover. We call this recovery step, a BGP routing
table transfer. Setting aside the table transfer delays, some routers
in large VPN networks are unable to keep up with the ever-increasing
number of VPN routes, which threatens the expansion of VPN provider
backbones and their ability to host new customers. In order to
lengthen the life of VPN backbones, prior work proposed incremental
solutions. Although these solutions proved helpful, we argue that they
fail to address the root cause of current scalability problems, namely
the original design of BGP and its inadequacy to the VPN routing
specific requirements. We show that BGP is not the best for what we
defined as VPN routing and propose a new approach for VPN routing.
The contributions of this thesis are threefold. The first is a passive
measurement study to understand the VPN route explosion phenomenon and
characterize it using routing data collected from a large VPN
provider. The second is a controlled experiment based study to explain
the reasons behind slow BGP routing table transfers. The results of
this study also apply to Internet backbones in general. The third
contribution is a new scalable approach for VPN routing.
==========
Jury
==========
- Jon Crowcroft, University of Cambridge (Rapporteur)
- James Roberts, INRIA, (Rapporteur)
- Christophe Diot, Technicolor (Examinateur)
- Sébastien Tixeuil, UPMC Sorbonnes Universités (Examinateur)
- Renata Teixeira, UPMC Sorbonnes Universités & CNRS (Directeur)
- Serge Fdida , UPMC Sorbonnes Universités (Co-directeur)
- Mickael Meulle, France Telecom R&D, Orange Labs (Encadrant)
==================
Accès
==================
LIP6 : 4 Place Jussieu 75005 Paris
Détails: http://www.lip6.fr/informations/comment.php
Cordialement,
Zied Ben Houidi
Aucun commentaire:
Enregistrer un commentaire