Abstract

In recent years, Internet has witnessed a rapid growth in P2P (peer-to-peer) applications, especially, in the live streaming domain. There have been several deployments of largescale industrial level P2P live video systems, e.g., CoolStream, PPLive, Sopcast. Several contemporary measurement studies have verified that thousands of users can simultaneously participate in these systems. Almost all live P2P video systems offer multiple channels (e.g., PPLive can host over 100 channels). It is expected that in near future, live streaming systems with hundreds of user-generated channels will likely have thousands of live channels in total. With such a large number of streaming channels and huge number of participants, there are still some challenging issues needed to be addressed for an efficient P2P live streaming system. This PhD thesis is organized around three such problems related to the P2P live streaming systems.
In the first research problem, our focus is on the Dedicated Channels used by a Small numbered Viewers (can be termed as DCSV channels for short) in a multi-channel live streaming system. Usually, these are user generated channels and they suffer adversely from poor channel performance, mainly, due to having a small number of participants. As a result, when a viewer of such a channel explicitly requests for a block of streaming content (commonly referred as a chunk), the probability that the chunk will be available among the existing viewers is less than it would be if the number of viewers was higher in that channel (e.g., viewers in a popular channel). We have proposed HnH (short for Hand-in-Hand), a novel scheme of cross-channel resource sharing, in order to solve the performance problem of DCSV channels due to their small number of viewers. We next develop a discrete-time stochastic model in order to analyze its efficiency.
In the second research problem, we focus on the Free riders who only want to download and watch the streaming content from their neighboring peers but are unwilling to upload any content to their neighbors. The presence of free riders impose obstacle to the stability of any live streaming system because of consuming bandwidth from the system without significant contribution. We have investigated the performance of a Live streaming system with and without the presence of a free riders. First, we develop a discrete-time stochastic model and then compare the probability of continuous playback without any free rider and with certain amount of free riders. Next, we introduce a simple incentive mechanism and modify our stochastic model in order to accommodate the incentive mechanism. Then we compare the result of probability of continuous playback with and without having an incentive mechanism. Our work shows that presence of an incentive mechanism improves the overall system performance.
In the third research problem, we focus on less motivated peers who are not interested to upload streaming contents to their neighbors if those contents are not from the channels they are watching. The context of this work is related to our first research problem where we have proposed the HnH scheme and for simplicity have considered that all the peers from all participating DCSV channels are motivated to cooperate. However, in practice, some of them may behave selfish and become less motivated to help peers from other channels. In this work, we investigate the performance of a HnH scheme based Live streaming system with and without the presence of less motivated peers. First, we develop a discrete-time stochastic model and then compare the probability of continuous playback without any selfish peer and with certain amount of selfish peers. Next, we introduce a simple incentive mechanism and modify our stochastic model accordingly. Finally, we compare the result of probability of continuous playback with and without incentive mechanism. Our work shows that presence of an incentive mechanism improves the over all system performance due to the fact that less motivated peers are motivated to cooperate more by the improved performance.