Impact of Location on Content Delivery

The increasing number of users as well as their demand for more and richer content has led to an exponential growth of Internet traffic for more than 15 years. In addition, new applications and use cases have changed the type of traffic. For example, social networking enables users to publish their own content. This user generated content is often published on popular sites such as YouTube, Twitter, and Facebook. Another example are the offerings of interactive and multi-media content by content providers, e. g., Google Maps or IPTV services. With the introduction of peer-to-peer (P2P) protocols in 1998 an even more radical change emerged because P2P protocols allow users to directly exchange large amounts of content: The peers transfer data without the need for an intermediary and often centralized server. However, as shown by recent studies Internet traffic is again dominated by HTTP, mostly at the expense of P2P. This traffic growth increases the demands on the infrastructure components that form the Internet, e. g., servers and routers. Moreover, most of the traffic is generated by a few very popular services. The enormous demand for such popular content cannot be satisfied by the traditional hosting model in which content is located on a single server. Instead, content providers need to scale up their delivery infrastructure, e. g., by using replication in large data centers or by buying service from content delivery infrastructures, e. g., Akamai or Limelight. Moreover, not only content providers have to copewith the demand: The network infrastructure also needs to be constantly upgraded to keep up with the growing demand for content. In this thesis we characterize the impact of content delivery on the network. We utilize data sets from both active and passive measurements. This allows us to cover a wide range of abstraction levels from a detailed protocol level view of several content delivery mechanisms to the high-level picture of identifying and mapping the content infrastructures that are hosting the most popular content. We find that caching content is still hard and that the user’s choice of DNS resolvers has a profound impact on the server selection mechanism of content distribution infrastructures. We propose Web content cartography to infer how content distribution infrastructures are deployed and what the role of different organizations in the Internet is. We conclude by putting our findings in the context of contemporary work and give recommendations on how to improve content delivery to all parties involved: users, Internet service providers, and content distribution infrastructures.