Routing and Control

To be translated real soon now

In this area, we are mainly interested in routing and switching in the network core, and their interactions with user activities. In a simplified model of Internet operations, users do not have direct influence on routing which is basically static in the absence of faults. This simple model has more and more exceptions. For example with multicast routing, receivers joining a group create routing entries in the core of the network. Similarly with multipath load sharing, changes in flows generated at the edges of the network may yield changes in routing in core routers. The study of this type of problems has some distributed algorithms aspects as well as protocol engineering aspects. One important evaluation criterion is scalability, since core routers have to deal with many flows and a limited processing capacity. One difficulty for the validation of this kind of study is the lack of precise and realistic netwwork models. This is why our work on this area may be currently divided into three topics: multicast routing, congestion control and modelisation for simulation and emulation.

Opérations de recherche

Multicast Routing

Participating Team Members : J.-J. Pansiot, D. Grad, V. Lucas, M. Hoerdt, C. Jelger

Multicast routing is a recognized activity of our team since several years [1]. This topic is now mainly aimed at congestion control and scalability of multicast routing, PIM-SM and its variants. In his thesis, M. Hoerdt[2] studied routing state repartition and proposed a new architecture, SSMSDP, allowing to use only source trees (SSM model, which is more robust than shared trees of the ASM model)while providing a dynamic multi-source model at the application layer. This work has been partly done in the framework of the RNRT @irs++ project, and the European project 6net. This architecture has also been proposed at the Ietf. More recently we have been working on the control and protection of the signaling plane of PIM-SM: Receiver join control to manage resources and mitigate Distributed Denial of Service against multicast routing. This is because multicast routing is one of a few internet mechanisms where end user actions trigger signaling inside the network. AS a result of this work (in collaboration with B. Hilt and M. Hoerdt) we have proposed a feedback channel for PIM-SM allowing to control useless multicast trees and DDoS attacks. In a work common with our mobility area, C. Jelger [3] and T. Noël have studied multicast communication with mobiles, particularly the difficult case of mobile multicast sources and tree reconfiguration.

Congestion Control and Load Sharing

Participating Team Members : J.-J. Pansiot, S. Cateloin, D. Grad, P. Mérindol, V. Lucas

This work has started more recently with S. Cateloin recruitment. We study (P. Mérindol Ph.D. thesis under the supervision of J.J. Pansiot and S. Cateloin) unicast load sharing among several paths towards the same destination. This requires constructing sufficiently many paths, not necessarily of equal cost. Moreover in a hop by hop routing model, loop avoidance is a hard problem. We have extended Dijkstra algorithm to find multiple paths and we have defined a routing mechanism taking the incoming interface into account and producing a larger set of lop free routes. The objective is to use multipath routing both for protection: locally switch traffic to a backup link in case of fault, and load sharing: locally and dynamically distribute flows on several paths. In his thesis V. Lucas(under the supervision J.J. Pansiot and D. Grad) is working on congestion control mechanisms for very large multicast groups, that is without receivers feedback to the source but using hierarchical layered coding. After some experimentations on a European platform we have discovered some problems with existing propositions such as webrc, and we are studying a new method based on a sliding window mechanism.

Network MOdels for Simulation and Emulation

Participating Team Members : J.-J. Pansiot, M. Hoerdt, D. Magoni, S. Cateloin, P. Mérindol

Our team has quite a long standing activity in Internet cartography since the study of Pansiot and Grad on routes and multicast trees in the Internet(ACM CCR Vol 28 N 1 1998)[4]. Network models have gained a big importance in our research community since many propositions are evaluated using simulation or emulation. In this context, unrealistic network models, for example topology models for routing, cannot give trustworthy results. We have continued studies on Internet topology including a cartography of the IPv6 enabled Internet with traceroute based techniques. Moire recently a study of the Mbone has been achieved using new tools, yielding more precise results on the local topology and its dynamics, but limited to multicast routers[5]. Beyond the topology itself, we are also interested by other parameters useful for different types of evaluation (delay, bandwidth, etraffic engineering metrics, traffic matrices, ...)