Mobile traffic is expected to grow tenfold by 2019, topping 24 exabytes of monthly traffic and accounting for nearly half of all Internet traffic. This growth is driven by the increasing number of smart phones and tablets, and the data demands of high bandwidth services enabled by next-generation cellular networks such as LTE/5G. As in the wired Internet, network usage is dominated by content consumption, with the vast majority served through content delivery networks (CDNs). CDNs host and replicate popular content across thousands of servers worldwide, directing users to “nearby” servers. This replica selection is a key determinant of client performance, yet replica selection for cellular clients has previously been overlooked, due to high radio latency, inconsistent throughput, and a limited number of ingress locations which dominated end-to-end latency. NGCNs and their improved performance place a renewed emphasis on replica selection policies for cellular clients. We find that the performance of existing replica selection systems in cellular networks is hindered by their opacity, the dynamic assignment of clients to infrastructure components, the emergence of centralized DNS within cellular networks, and the growth of public DNS in global mobile operators. This opacity prohibits network probes from entering these networks, rendering existing monitoring and measurement systems ineffective. In this dissertation, I argue for the centrality of cellular network packet gateways (PGWs), and that this centrality has critical implications on the architecture, characterization, and performance of cellular networks. PGWs separate the interior mobile network from external data networks, and define the independent network partitions which compose modern cellular networks. I posit that understanding the locations of PGWs and their allocation of clients constitutes sufficient network topology coverage. The presence of PGW’s on all routes to and from cellular clients make them ideal proxies of client latency for network services. I demonstrate techniques for characterizing cellular networks which allow both the discovery of PGW locations and their assignments of mobile clients. I designed and implemented two live systems which utilize these techniques to characterize cellular infrastructure: tiller which uses instrumented mobile devices to characterize cellular networks, and machete which uses traces from external vantage points to accomplish this characterization at a global scale. I introduce a novel method of content replica selection which chooses cellular client servers based on the location of a client’s PGW, called Gateway-Based Replica Selection (GBRS), and show this achieves near optimal replica selection for cellular clients.

Last modified

• 03/28/2018

Creator

• John Paul Rula

DOI

• https://doi.org/10.21985/N26Q44

Subject

• Computer Science

Keyword

• Mobile Network Measurement
• Content Delivery Networks
• Cellular Networks

Date created

• 2017-01-01

Resource type

• Dissertation

Rights statement

• In Copyright