UC Santa Cruz

Extensive growth of Internet applications like communication, education, and leisure, along with the privacy concerns and significant amount of personalized contents, have clearly affected the connectivity of the networks and the users' behavior. The social nature of today's networks' users (the term 'social' is used as a general technical term for network users with specific mutual relationships, and can address a large domain of users, including but not limited to social media users) has in turn a major influence on the networks' shapes and characteristics. On the other hand, with the emergence of advanced hardware and software technologies which enable significant processing power and storage space in mobile devices, and with their ever increasing widespread use, today's Internet is moving from an infrastructure-based network towards a wireless ad-hoc network. Besides, caching data within the network is going to be inevitable to improve latency and reduce bandwidth consumption, and storage is being considered as one of the network primitives.

This research, thus, investigates the impact of the above aspects of today's networks on the performance. It first discusses and evaluates the effect of users' social behavior on the maximum achievable data rate in the wireless ad-hoc networks, and proves that social connection among nodes may actually help in scaling wireless networks. It also reveals that due to their different social status, various users have different effects on the performance, and therefore, traditional transport capacity concept for wireless networks is not appropriate for these types of networks.

Second part of this work investigates the improvements in the fundamental limits of the performance metrics in networks of caches, and evaluates the effect of different caching policies and content searching algorithms on the obtained improvements. Then a framework is presented to quantify the overhead traffic of locating contents, and is later used to define some optimal policies with respect to the contents that should be cached for an operator-driven content distribution system. This framework can generally be used in many other distributed systems contexts where a control plane has to stay aware of the state of the forwarding plane.