Lieu : salle des Thèses B312 à Télécom ParisTech (46, rue Barrault – 75013 Paris)
Membres du jury :
Rapporteurs :
- Subhabrata Sen (AT&T Labs Research)
- Reza Rejaie (University of Oregon)
- Kurt Tushku (University of Wien)
Examinateurs :
- Serge Fdida (LIP6)
- Daniel Kofman (Télécom ParisTech)
- Walid Debbous (INRIA)
- Marco Ajmone Marsan (Politecnico di Torino)
Titre : “Peer-to-peer networking: The Dawn of the Homo Interconnexus Era”
If we take into account the importance that telecommunication networks have nowadays reached, we might be thinking that, in the long Darwinian evolution of mankind from “Homo abilis” to “Homo sapiens”, the next step will likely be that of “Homo interconnexus”. Peer-to-peer (P2P) communications represent a crucial face of the above recent evolution of Internet user habits: besides the ever-present file-sharing applications as BitTorrent and eMule, we use P2P applications such as Skype to call our friends with VoIP; for entertainment purposes, we rely on P2P-VoD and live TV applications such as SopCast and PPLive; moreover, even operating system with and applications are moving toward P2P distribution of their updates, of which aptp2p and Blizzard are respective examples.
This HDR memoir covers a few important aspects of P2P networking, and especially its interaction with the underlaying IP networks. One research line goes along the effort to understand how popular P2P applications, such as Skype, PPLive and SopCast, work, and to characterize the user behavior. As these applications are typically closed-source, implement unknown proprietary algorithm and extensively use encryption techniques, understanding their inner working is very often a challenge.
As P2P services generate traffic that is ultimately transported by the IP infrastructure, a second research axis addressed in this memoir focuses on the study of the impact of P2P traffic on the underlaying network: with this respect, we studied not only the spatial properties of P2P traffic (e.g., network awareness), but also from a congestion control standpoint. Our research shows striking contrasts of aggressive P2P-TV and P2P-VoIP applications (that need to guarantee their users with high-quality interactive content) with more gentle P2P file-sharing (that can put itself in low-priority and yield to TCP and other traffic) and helped designing such new protocols.
Finally, instrumental to both research lines above, is the ability to classify the traffic (i.e., recognizing which application has generated the traffic under observation), which is also crucial for network operators, as it enables differential treatment of packets for QoS, tariff purposes, legal interception, etc. In this context, our research introduces novel statistical description of either the syntax or the behavior of P2P applications, both inspired by human communication principles.
