Impact of Delivery Eco-system Variability and Diversity on Internet Video Quality

We study the impact of the variability and diversity of multiple Internet delivery eco-system components (CDNs, ISPs, devices, video players, streaming protocols) on the quality of Internet video. Our study is based on a data set that is unique in two aspects: (a) client-side instrumentation: this provides critical insights into video quality as the video has traversed all eco-system components before reaching the client; (b) large-scale: our data set spans scores of popular sites, billions of streams, hundreds of thousands of video objects, and hundreds of millions of unique viewers. We show that there is large variability and diversity in each of the many Internet video delivery eco-system components, which can significantly impact the quality of video that is delivered to the end users. We present a set of architectural principles that are essential to optimize video quality in the presence of variability and diversity.