Nick Feamster - Investigating Pricing Models in Internet Connectivity Markets

Content providers, such as e-commerce sites and media outlets, use content delivery networks (CDNs) to deliver their content with better performance to their end-users. In return for their investment in better performance, content providers expect changes in user behavior that lead to higher revenues and profitability. An e-commerce provider expects that the faster loading of web pages will eventually lead e-shoppers to shop more on the site. A media provider expects that videos that startup quickly and play without interruptions will result in viewers abandoning less, viewing more, and coming back more often, enhancing ad revenues and customer loyalty. But, can better network performance cause significant changes in user behavior? While there are many small-scale studies and much correlational evidence, this critical question leaves much to be explored from a scientific standpoint. We argue that the time is finally ripe for truly large-scale scientific studies that explore causal relationships between network performance and user behavior. Such research will strongly influence the future direction of both network architecture and the economics of the content distribution. Our recent work in inferring causal relationships between CDN performance and viewer behavior for streaming media is a step in this direction. Using a novel Quasi-Experimental Design (QED) technique, we show that a user who experiences a rebuffer delay equal to 1% of the video duration plays 5% less of the video in comparison to a similar user who experienced no rebuffering. Further, we show that users start to abandon a video if it takes more than 2 seconds to start up, with each incremental delay of 1 second resulting in a 5.8% increase in the abandonment rate. Finally, we show that a user who experiences failure is 2.3% less likely to revisit the same site within a week than a similar user who did not experience a failure.