Measurement of Commercial Peer-to-Peer Live Video Streaming
(2006)Abstract
Peer-to-peer technologies have proved to be
effective for various bandwidth intensive, large scale applications
such as file-transfer. For many years, there has
been tremendous interest in academic environments for live
video streaming as another application of P2P. Recently, a
number of new commercial scale video streaming systems
have cropped up. These systems differ from others in the
type of content that they provide and attract a large number
of users from across the globe. These are proprietary
systems and very little is known about their architecture
and behavior. This study is one of the first of its kind to
analyze the performance and characteristics of P2P live
streaming applications. In particular, we analyze PPLive
and SOPCast, two of the most popular systems in this
class. In this paper, we (1) present a framework in which
to analyze these P2P applications from a single observable
point, (2) analyze control traffic to present a probable
operation model and (3) present analysis of resource usage,
locality and stability of data distribution. We conclude that
P2P live streaming has an even greater impact on network
bandwidth utilization and control than P2P file transfer
applications.
Description
First, we
separated the control traffic from the data traffic in the
collected traces. This allowed us to analyze the control
traffic separately and reverse engineer the protocols used
to get an understanding of how these systems work. We
present an overview of how these systems operate based
on our analysis of the control traffic. Next, we define
a generic framework that can be used to evaluate the
data distribution performance of such systems. For that
purpose, we define metrics to highlight the key characteristics
of the distribution plane. Finally, we evaluate the
two systems using the framework formulated and present
results.