Research Areas
I am currently working on Peer-to-peer video streaming with Prof. Bernd Girod.
Introduction to peer-to-peer multimedia streaming systems
Since the success of peer-to-peer file sharing systems such as Napster and GNUtella, the research community has considered using peer-to-peer networks to replace the traditional client-server paradigm for video streaming. However, providing this application layer multicast service is difficult due to the dynamic nature of the host computers (peers) which hampers scalability and stability.
In peer-to-peer networks, peers contribute their upstream bandwidth to the system. The fact that a peer serves other nodes can cause service disconnection when the peer leaves. To minimize the service disruption due to peer departure, we propose a P2P streaming system based on the overlay of multiple trees. To exploit path diversity, we use other trees to request a missing video packet when one of parents leaves the group. A small number of fan-out of a peer node in constrast to a large fan-out of a media server will enable us in the future to locally perform packet scheduling based on a packet's importance such as the aggregate contribution to video quality across the descendant peers.
The characteristics of peer-to-peer multimedia streaming
-
peers, or ordinary end host users like home computer users or laptop users in the public areas participate in watching and relaying video to other peers.
-
peers are inherently unreliable compared to the network infrastructure in a sense that they can join and leave the live session anytime.
-
video and/or audio delivery requires low latency property. Data dissemination or large-scale distributed systems have little delay constraints.
-
streaming requires real-time streaming. Part of the video may be dropped and not transmitted to receivers. On the contrary, data dissemination or large-scale distributed systems require 100% data reception.
Our research goals
We design a new peer-to-peer multimedia streaming system by understanding the difference between data distribution and multimedia streaming, in order to achieve a robust, low-latency, high-quality video quality among users.
Using multiple video dissemination paths(a.k.a. trees) makes the system more robust against single points of failure. However, it also creates more parent failure and potential video quality degradation. This paradox of improving/huring video quality with multiple trees is one of the cruicial issues in our research.
References
-
Ganjam, A., Zhang, H. "Internet multicast video delivery," Proceeding of IEEE, January 2005.
-
Yang-Hua Chu, Sanjay G. Rao, Srinivasan Seshan, Hui Zhang, "A Case for End System Multicast," IEEE Journal on Selected Areas in Communications, Volume 20, Issue 8, Oct. 2002.