Committee: Reza Rejaie (chair), Allen Malony, Jun Li
Directed Research Project(Sep 2015)
Keywords: stub-AS, traffic locality, network measurement
A rich and very dynamic Content distribution ecosystem enables today's Internet users to access content associated with major providers from a close-by front-end server. While the advantages of the resulting locality of traffic for users (i.e. better performance) and providers (i.e. lower cost) are well understood, two basic questions about the content distribution ecosystem remained unanswered (i) What is the typical level of traffic locality for users at the edge of the network? and (ii) Whether the traffic locality affects the observed bandwidth by these users? This paper presents an empirical assessment of traffic locality for a stub-AS to answer these questions. We employ unsampled Netflow data from University of Oregon's (UONet) border gateways to assess the level of locality for top content providers of this network using four metrics of distance, namely geographic distance, router hop, AS hop and round-trip time (RTT). We also present a method to identify front-end servers of a major content provider, namely Akamai, that are placed within other ASes with the intention to increase their traffic locality (called guest servers).Using this method, we identify Akamai's guest servers that serve UONet. Finally, we examine the effect of traffic locality on the observed performance (i.e. download bandwidth) by UOnet users and show that traffic locality does not have a strong correlation with the download bandwidth of UOnet users.