In parallel, with Dali
Kaafar and Walid
Dabbous, we have designed a scalable locating algorithm
called LCC
that
reduces the convergence time of the multicast overlay construction.
This algorithm has been implemented in a library that will be soon
available in the public domain along with wrappers for various
multimedia applications (vic,
rat,
vlc). We
have also studied how to
extend it to mobile ad hoc environments.
I had the great opportunity to work on Software Radio since my postdoc done at MIT in 95-96 with the SpectrumWare project (a pioneering project in this area) headed by David Tennenhouse. There, I have designed and implemented in general-purpose computers, part of a GSM base station which has been integrated into the VuSystem (preceding GNU radio). I also proposed to use a new metric to estimate complexity of such greedy-CPU algorithms.
Back to INRIA, I continued to work on this exciting new area within the Hahnsang Kim, we developed the EPspectra toolkit based on the Esterel formal language and the PSPECTRA toolkit developed within the SpectrumWare project.
Then, my work focused on how to take advantage of software radio, in particular, because it is now possible to adapt all the layers of the communication protocol stack (including the PHY layer), how to design efficient cross-layer mechanisms. For example, with Hossein Manshaei, we designed an algorithm to select the PHY mode of 802.11 stations that takes into account both the wireless channel characteristics and the intrinsic characteristics of the application (e.g. for video streaming, the loss tolerance of the video decoder).
Once, with Anwar Al Hamra and Chadi Barakat, we have studied the possible benefit of network coding for wireless mesh networks in a file sharing context. This approach seemed promising…
IEEE 802.11a/b/g have various PHY transmission modes, and these modes (with different PHY modulations, rates and FEC levels) should be set according to the channel characteristics. Because the wireless medium is very variable, such mode selection mechanisms are tricky to perform. With Hossein Manshaei, Mathieu Lacage and Ceilidh Hoffmann, we have proposed different algorithms. Some of them have been implemented within MadWifi.
Transmission of real-time multimedia flows over IEEE 802.11 networks is problematic, especially when the wireless medium is congested. Service differentiation mechanisms are proposed in the upcoming IEEE 802.11e standard but this is not panacea. With Qiang Ni, we proposed several solutions to improve the new HCCA and EDCA services. I also designed with Jean-Baptiste Lapeyrie a new scheduling algorithm that supports service differentiation in Bluetooth Piconets.
I had a few contributions on network tomography, especially concerning the MINC tool with my PhD student Vijay Arya – actually he was the main investigator of these works. Basically, we have designed mechanisms to make MINC more robust against possible misbehaving nodes and other mechanisms to reduce the amount of feedback sent.
Large Scale Virtual Environments (LSVE) require very scalable communication protocols. I started working on this area in 1998 with Emmanuel Léty (my first PhD student). Together and with the help of Francois Baccelli, we designed SCORE, a scalable communication protocol that dynamically partitions the virtual environment into spatial areas and the association of these areas with multicast groups. SCORE has been experimented within the V-Eye LSVE application. Then, with Walid Dabbous and Laurentiu Barza, we proposed another communication protocol called SCORE-SSM which assumes the Single Source Multicast (SSM) model as the underlying network infrastructure.
I started working on congestion control in 1992, while developing IVS. This pioneering application allowed me to evaluate new congestion control algorithms in real conditions. In a nutshell, in best effort IP networks, since there is no guarantee of packet delivery (packet loss), delay and jitter, the application has to be adaptive. In particular, multimedia applications have to implement their own congestion control algorithms to adapt their transmission rate to the variable available bandwidth. With Jean Bolot and Ian Wakeman, we have designed one of the first multicast congestion control mechanisms for real-time videoconferencing over the Internet.
With Jean, we designed several flavours of such congestion control algorithms including TCP-friendly algorithms. I continued this work in collaboration with Christine Guillemot and Kavé Salamatian, focusing on how to improve these mechanisms in the presence of multiple heterogeneous receivers.
When I was PhD student in the High-Speed Networking Research Group RODEO, I started working on adaptive multimedia applications over the Internet. Specifically, I developed IVS, one of the first videoconferencing application for the Internet. At that time (91-92), most of people believe that 1/ real-time video could not be displayed on workstation's screens, 2/ workstations could not implement efficiently software codecs and 3/ the Internet could not be used to transmit real time multimedia...
I implemented in software a H.261 video codec on general purpose workstations (Sun, HP and decstations). I also designed specific error control and rate control mechanisms for transmission of video streams and a packetization algorithm that has been standardized by the IETF [RFC-2032]. These research activities have been done in particular with Jean Bolot and Christian Huitema (my PhD supervisor).