Content dissemination for static and dynamic mobile P2P groups

In this thesis we present content dissemination protocols for static and dynamic peer-to-peer (P2P) groups. In the first part of the thesis we delve into the problem of inefficient content dissemination for static mobile device users due to cellular congestion. We propose the Sangam Framework which utilizes multiple radio interfaces and efficient scheduling policies based on device resources. In the second part of the thesis we explore solutions for efficient content dissemination for dynamic P2P groups. In order to obtain more information about mobile users, we analyze user mobility traces and attempt to construct a social P2P network. In the third part of the thesis we combine the Sangam framework and the mobility analysis in order to design the dynamic group based content distribution framework. Solutions to the cellular network bandwidth problem have been presented by the community such as usage of alternative wireless (e.g., 802.11 WiFi) network and P2P file sharing over a group of wireless devices. However, often the theoretical and simulation approaches for file sharing within multi-radio P2P groups hide the complexity of systems and networks in real scenarios such as heterogeneity of phones in a P2P group, issues with scheduling policies within a group of devices, group formation, etc. In the first part of the thesis, we present Sangam (``confluence'' in Sanskrit), an efficient cellular-WiFi group framework for file sharing where we address extensively system and network challenges in file sharing for real phone group scenarios. The Sangam framework accounts in its protocol, policy and algorithmic designs for (a) heterogeneity of phone group devices in terms of CPU and power levels, (b) different sizes and numbers of chunks in P2P part of the group-based content distribution, (c) hybrid scheduling policies for chunk dissemination within multi-radio wireless group environment, and (d) different group sizes. Sangam validation shows the impact and difference for simulations when considering real implementation of video file sharing within a cellular-WiFi group of Android phones. In the second part of the thesis, we attempt to explore user mobility and group formation. Smartphone users have created large virtual and physical mobile communities by virtue of their physical location and/or social relationship with other smartphone users. Social scientists observe, log via surveys, and classify these smartphone mobile communities according to their social interactions. Some social scientists also consider coarse physical features to develop socio-physical models. On the other hand, computer scientists observe, log low-level fine-grained physical features, and classify these communities according to their physical spatial and temporal measurements. However these detailed physical measurements are not well reflected in existing socio-physical models. We present the physical refinement of the existing socio-physical model for Situated Wireless Communities (SWC) using physical phone traces of the Mobile Learning Community (MLC) at the University of Illinois at Urbana-Champaign. We take the bottom-up approach where through data mining and inference techniques, we analyze and infer from low-level physical phone features new higher-level social group behaviors such as Physical Stable Temporal, Physical Stable Spatial, Physical Dynamic Temporal and Physical Dynamic Spatial behavior. Our results show that the MLC groups exhibit new social behavior, in physical spaces across temporal and spatial dimensions which were not identified in the existing SWC socio-physical model. Our refinement approach also shows the methodology using which one can identify new socio-physical group behavior as more and more detailed physical phone features are being collected. In the third and final part of the thesis, we will present the Dynamic Group Based Content Distribution Framework which combines the content distribution framework presented in part one and mobility analysis of users in part two. Initial P2P networks like Gnutella and Napster targeted scenarios where users were interested in popular media files. There was no social aspect of users that was considered. It was a need based network. Users willing to share the file were contacted, and based on the link quality the file was downloaded by the interested user. The overlay links did not consider social relationships, clustering of users with common preferences, etc. With the evolution of the internet and its applications, social networking has become popular. The social graph that can be constructed, using data from social networking sites, provides a wealth of information. P2P networks like Socionet and Prometheus use this information to construct smarter overlays enabling users to query and fetch items more efficiently. With the large number of mobile phone users, P2P networks provide more advantages in terms of bandwidth, power and download time efficiency. Using the mobility traces collected by the University of Illinois Movement (UIM) tool and the Sangam framework, the Group Based Content Distribution Framework groups different mobile users together at different locations. Using metrics such as average duration of time spent by a user at a given location, overhead of peer membership, distinct file segments offered by the peer, etc., for P2P links, a central server groups mobile users and schedules distinct file segments. The evaluation of the Dynamic Group Based Content Distribution Framework shows the bandwidth overhead and increased download time of peers in the absence of the Dynamic Group Based Content Distribution Framework