Offering Interactivity and Reliability for Peer-to-Peer Video-on-Demand

PhD Thesis Proposal Defence


Title: "Offering Interactivity and Reliability
for Peer-to-Peer Video-on-Demand"

by

Mr. Wai-Pun Yiu


Abstract:

Peer-to-peer (P2P) technologies provide a promising solution for the
development of attractive networked multimedia applications like
video-on-demand (VoD). Relatively speaking, there has been not much work
on providing VoD service using P2P approaches. Providing P2P on-demand
streaming is challenging due to unreliability of end-hosts and
difficulty in locating supplying peers, which we address in this
proposal. In this proposal, two critical issues in P2P VoD systems are
studied and addressed. The first issue is to locate supplying peers who
possess the requested media data in a large P2P network. This is
important for supporting VoD user interactivity functionalities such as
random forward/backward seek. After supplying peers are located, the
second issue is to effectively and efficiently distribute data among
peers. In this context, scalable error recovery techniques are required
to provide resilient service.

We first propose a scalable P2P VoD system named PRIME, which is based
on sliding window buffering at client sides. We introduce the concept of
virtual arrival slot and apply distributed hash table (DHT) for locating
qualified parents during startup and random seeking. Preliminary results
show that PRIME achieves low startup and seeking delay. However, like
other sliding window buffering approaches, missing packets due to packet
loss and node failure are accumulated along data paths. Therefore, we
present a novel error recovery scheme called LER to enhance the error
resilience of media streaming systems. In LER, peers are assigned to
different planes, on each of which peers form a streaming overlay.
Packet loss is recovered from peers on other planes rather than those on
the same plane. This scheme is expected to be better than traditional
parent recovery because the errors experienced by a peer and its parents
are highly correlated. In PRIME, there are more suppliers for more
popular video segments, thus it is intrinsically popularity-aware.
However, its cache-and-relay approach makes its peers suffer from buffer
shortage during parent interactivity like random seeking. This is
because the parent may start caching media data which is no longer
interested by the child peer. Hence, we introduce another system called
VMesh, in which peers statically store video segments at local storage
and serve others using the stored segments. We will show that
VMesh would achieve a good performance under lossy network condition and
random user interactivity. Since segment popularities (or access
probabilities) are usually non-uniform, we will study a popularity-aware
segment distribution protocol to allocate segments dynamically according
to their popularities.


Date:     		Tuesday, 6 November 2007

Time:                   10:30a.m.-12:30p.m.

Venue:                  Room 3402
			lifts 17-18

Committee Members:      Dr. Gary Chan (Supervisor)
                        Prof. Mounir Hamdi (Chairperson)
			Dr. Jogesh Muppala
			Dr. Oscar Au (ECE)


**** ALL are Welcome ****