UC Riverside

The focus of this dissertation is on measuring, analyzing and modeling emerging appli-

cations in the Internet. Specifically, we concentrate on understanding the internals of content

distribution paradigms such as Peer-to-Peer (P2P) systems and podcasts. This dissertation

consists of three main thrusts which we describe below.

P2P streams have been reported to constitute nearly 61% of all upstream traffic. P2P

streams are used for disseminating content ranging from video programs to linux images.

This everpresent ubiquity of P2P networks has also allowed them to be used for sharing

copyrighted material. This has resulted in organizations like the RIAA, taking legal action

against file-sharers. As a result P2P users have employed defenses against being monitored

by such organizations. We have found that a little caution pays off a lot, since there is a 100%

probability of a naive P2P user being monitored when accessing these networks.

Further, as a case study, we present a comprehensive study about eDonkey, a popular

P2P network. We identify the limitations of current approaches to measure P2P networks.

Additionally, we find that P2P flows traverse through the Internet quite differently than http

flows. Based on this, we present metrics useful for distinguishing P2P traffic from other

forms of traditional content distribution in the Internet.

Finally, podcasts, a relatively new content distribution mechanism is expected to garner

an audience of nearly 56 million subscribers by 2010. Measuring and modeling podcasts re-

mains an open problem despite the significance that has been gained by this application. This

form of content distribution is best described as a push based mechanism, which is different

from traditional http based content distribution. We measure podcast streams, analyze them

and develop a traffic generator, SimPod, for simulation purposes.