SCS

Abstract

In several situations it is desirable to distribute content to all users in a wireless network. Examples include the dissemination of software and firmware updates, push-based video-on-demand systems, and news feeds. When the content is not extremely time-sensitive, there is an opportunity to reduce the consumption of server-client bandwidth by exploiting local connections between users. Rather than transmitting the entire content to each user individually, the content is divided between multiple users. When users meet they exchange data, with the result that the content percolates through the network and eventually each user has the entire content. As a network designer, we have no control over the contacts between users, which are dictated by social interactions and user behaviour. We can model the contact events as a random process and, making certain assumptions about the behaviour of this process, design procedures to optimally distribute the content among the users so that it spreads through the network as quickly as possible. In this talk, I will describe a content distribution procedure based on clustering the users into groups using a network clustering algorithm that strives to maximize the conductance of each cluster. The procedure is suboptimal, but the performance deterioration is relatively small, and there are significant improvements in terms of estimation overhead and robustness.