Diego Perino, Orange Labs

Location :

Orange Labs - room S012
38-40, rue du General Leclerc
92794 Issy-les-Moulineaux

Please register in advance! 
Contact Fabien Mathieu (fabien.mathieu@orange-ftgroup.com) or Diego Perino (diego.perino@gmail.com) to announce that you are coming. 

Abstract :

Multimedia streaming over the Internet strongly increased its popularity in last years.  
Media traffic has kept growing and is expected to increase tenfold within five years.
To support such big audience, the traditional server-based architecture requires huge amount of storage, bandwidth and computational resources. This thesis considers streaming applications based on peer-to-peer architectures, which may overcome these resource constraints and cope with these evolutionary trends. In a peer-to-peer system the resources increase with the number of users; we show this is indeed effective to improve service scalability while reducing provider costs. 
 
A first part of the thesis is devoted to the mesh approach to peer-to-peer live streaming, which has been used by popular commercial applications like PPLive and SoapCast. Through an experimental evaluation of PULSE, an unstructured peer-to-peer live streaming system we designed and developed, we analyze the streaming process in unstructured systems, and the impact of locality and resource awareness on their performance.
To better understand the critical aspects of the stream dissemination in unstructured networks we focus on the building blocks of the diffusion process: the chunk/peer selection algorithms. We design and analyze some simple, yet practically interesting, selection policies  for homogeneous and heterogeneous systems, and we consider the impact of system parameters such as the source selection policy, the chunk and neighborhood size. Our approach mixes theoretical results, when available, with empirical observations in order to give the best possible insights. 

A second part of this thesis considers the on demand streaming, and in particular server-free approaches where clients collaborate to store and distribute contents to serve requests generated by other users. We analyze the size of the content catalog such kind of systems can provide to their users, as a function of storage and bandwidth constraints. By means of simulations and experimental evaluations, we analyze simple practical techniques that can be used for content storage and distribution in a fully peer-to-peer on-demand streaming system.



Committee

Advisors:
Fabien Mathieu, Orange Labs
Laurent Viennot, INRIA

Referees: 
Marco Ajmone Marsan, Politecnico di Torino
Pascal Felber, Université de Neuchatel
Anne-Marie Kermarrec, INRIA

Examiners:
Pierre Fraignaud, CNRS
Arnaud Legout, INRIA
Laurent Massoulié, Thomson Lab

Host :

Fabien Mathieu and Diego Perino