Internet Performance: Who Cares and by How Much?

Internet services are at the core of human activity today. Billions of users interact with networks of various kinds for commerce, entertainment, news, and social networking. But, in the late 1990’s, things were different. E-commerce was in its infancy. Streaming video was an euphemism for audio, as Internet radio dominated the scene. Social networks didn’t exist. And, at Akamai, we were creating one of the world’s first content delivery networks (CDNs) [3]. All networks, including CDNs, are designed with a notion of performance in mind, a notion commonly referred to as quality-of-service (QoS). For a content delivery service, good QoS means web pages download faster and videos start up quickly and play without rebuffers (i.e., freezes). As computer scientists, we understand how to architect networks to provide good QoS. After all much of computer science is a study of performance in one form or another. But, we do not have a good scientific grasp of the relationships between QoS, the more subjective notion of quality of experience (QoE), and the causal impact of QoS on QoE, user behavior, and economic outcomes. In this paper, I argue that the time is ripe for unravelling these relationships, a research agenda that could lead to vastly better network architectures, deeper knowledge of user behavior, better monetization, and more informed Internet policies and governance.