Today, the Internet is increasingly becoming a platform for online services and massive content distribution. However, it was originally designed to primarily support pairwise host-to-host communications and thus is poorly suited for content dissemination among multiple hosts. With voluminous video content becoming the dominant traffic, such host-centric paradigm is increasingly revealing its inefficiencies. In recent years, we have seen a variety of attempts to provide more efficient content delivery support on the Internet, including i) research efforts of content-centric networking (CCN) as a clean-slate redesign of the Internet architecture; ii) the wide adoption of CDN services; and iii) the fast expansion of cloud infrastructures offering vital support for modern data-intensive ap...[ Read more ]

Today, the Internet is increasingly becoming a platform for online services and massive content distribution. However, it was originally designed to primarily support pairwise host-to-host communications and thus is poorly suited for content dissemination among multiple hosts. With voluminous video content becoming the dominant traffic, such host-centric paradigm is increasingly revealing its inefficiencies. In recent years, we have seen a variety of attempts to provide more efficient content delivery support on the Internet, including i) research efforts of content-centric networking (CCN) as a clean-slate redesign of the Internet architecture; ii) the wide adoption of CDN services; and iii) the fast expansion of cloud infrastructures offering vital support for modern data-intensive applications. We coin “content-oriented networking” to mean all these networks and services that make the Internet become a better platform for content distribution. In this thesis, we address several important practical problems on resource management and optimization in such content-oriented networks.

This thesis consists of three parts. The first part is about content management in CCN. CCN adopts pervasive in-network caching. But the default ubiquitous-LRU caching scheme leads to serious on-path redundancy. We begin with proposing a new caching scheme PCP and demonstrate via trace-driven simulations that it outperforms ubiquitous-LRU in both microscopic and macroscopic views. PCP and many other newly proposed caching schemes in the literature focus on reducing on-path cache redundancy, which relies on either explicit or implicit cooperation between nodes along the request-response path. This approach inherently suffers from the scalability issue and the hurdle of requiring uniform policy enforcement across different autonomous systems (ASes). To overcome these limitations, we next propose an intra-AS cache cooperation scheme iCCS, which commits to eliminating both on-path and off-path cache redundancy within a specific AS via periodic cooperative redundancy eliminations. Through trace-based simulations on multiple realistic network topologies, we show that iCCS significantly shortens the request-response latency, and dramatically reduces the amount of transit traffc without increasing internal link congestions. After that, we study the content peering problem in CCN by proposing an optimization model CPP. With extensive numerical experiments under realistic AS-level peering graphs, we find that the interconnectivity of the peering graph greatly influences the maximum potential peering benefit, and that cooperative caching could yield higher peering benefit than local greedy caching but is sensitive to various parameters.

The second part is dedicated to content multi-homing with multiple CDNs. We propose the MCDN-CM model, aiming to make an optimal operating plan for an authoritative CDN by deciding how to supply content requests with guaranteed QoE. We show via numerical experiments under realistic settings that MCDN-CM achieves a tremendous cost-saving compared to other methods.

The third part is focused on SDN-based traffic management in inter-data center networks. We propose the MCTEQ model, which adopts utility-based joint bandwidth allocation for multiple classes of traffic and in particular provides end-to-end delay guarantee for interactive flows. We demonstrate via numerical experiments that MCTEQ achieves considerably higher network utilization than the best existing solutions, while running at least twice faster.