Resource Allocation in Operator-owned Content Delivery Systems

Abstract: Live and on-demand video content have become the most important source of network traffic in mobile and fixed networks in recent years. In order to be able to efficiently deliver the increasing amount of video content, network operators have started to deploy caches and operator-owned CDNs. These solutions do not only reduce the amount of transit traffic of the operators but they may also improve the customers' quality of experience, through bringing the video content closer to customers. Nevertheless, their efficiency is determined by the algorithms and protocols used to allocate resources, both in terms of storage and bandwidth. The work in this thesis proposes and analyses algorithms for the allocation of these two resources for operator-owned content management systems. In the first part of the thesis we consider a cache maintained by a single network operator. We formulate the problem of content caching in a mobile backhaul as an integer program and we show that there exists an efficient centralized solution to the problem. Due to the prohibitive space complexity of the centralized algorithm, we propose two distributed approximations based on local information on the content demands. We then consider the problem of managing cache bandwidth so as to minimize the traffic cost of content delivery and we propose various approximations of the optimal stationary policy. We then consider the interaction among content management systems maintained by different network operators. First, we consider the problem of selfish replication on a graph as a model of network operators that use their caches to prefetch popular content, and try to leverage their peering agreements so as to minimize the traffic through their transit providers. We design efficient distributed algorithms that compute a stable content allocation through selfish myopic updates of content allocations at different network operators. We show that, if the cost function is neighbor-specific, network operators need bilateral payments to compute a stable content allocation that is individually rational. We then consider the problem of coordinated caching in a network of autonomous systems engaged in content-level peering. We investigate whether interacting operator-owned caches need explicit coordination in order to reach a stable content allocation efficiently. Beyond the theoretical contributions made to the analysis of player-specific graphical congestion games and their generalizations, the results in thesis provide guidelines for the design of protocols for standalone and for interconnected operator-owned content management systems.