Impact of Network Effects on Application Quality

Abstract

The ubiquity of high speed Internet access, proliferation in the adoption of mobile devices, and the popularity of content-rich applications have brought a new dimension to the Internet landscape. Indeed, high speed residential connectivity and mobile wireless access have changed user expectations. The broadband access technologies (i. e., DSL, WiFi, 3/4G, and LTE), and smart mobile devices (i. e., Android, iPhone, iPad, etc.), have enabled users to interactively browse, stream videos (i. e., YouTube and Netflix), play online games, and share content for social networking. All trends together have caused a fundamental change in how users interact with the Internet. Any adverse impact due to high Internet traffic, heterogeneous access, and application protocol mix on flows of different applications can result in sub par network performance and unsatisfactory user experience. Understanding the relation between network performance and user perception is thus crucial for application designers, network, and service providers. In this thesis, we endeavor to explore the impact of emerging network effects on different applications both from network performance and user experience point of view. Studies of network performance and user experience require a multi-purpose heterogeneous testbed that supports a variety of networking conditions commonly present in today’s Internet. We propose the design and architecture of QoE-Lab. The main features of QoE-Lab include 1) Next Generation Mobile Networks (NGMN), i. e., WiFi and 3G UMTS, 2) access/backbone network emulation, and 3) virtualization. It provides services like traffic generation, topology emulation, and high-precision cross-layer monitoring. We describe two Quality of Experience (QoE) case studies to show the benefits of the QoE-Lab testbed framework. Next, we perform a sensitivity study of the packet loss process within a router for different network load levels, flow size distributions, and buffer sizes. We compare the loss process for TCP and UDP flows at different link utilizations and buffer sizes. We highlight the importance of understanding the flow-level properties of the traffic, e.g., packet loss under different networking conditions and their consequences on application performance, i. e., flow-happiness. We find that packet losses do not affect all flows similarly. Depending upon the network load and the buffer sizes, some flows either suffer significantly more drops or significantly less drops than the average loss rate. Based on anonymized packet level traces from more than 20,000 DSL lines, server logs from a large content distribution network (CDN), and publicly available backbone traces, we investigate the flow-level performance of popular applications across a range of size-based flow classes. We use retransmissions, throughput, and round-trip-times (RTTs) as key flow performance metrics. We compare these metrics under different network loads, DSL link capacities, and for up/downstream directions. We show that irrespective of the direction, flows are severely impacted by events related to network load and application behavior. We also find that, in general, this impact (as measured by our performance metrics) differs markedly across the different flow classes. In particular, contrary to popular belief, small Thesis: “Impact of Network Effects on Application Quality”