Understanding data-center driven content distribution

Abstract

Cloud services are radically changing the way we access the information from the Internet and how the "digital goods" are delivered to users. To enable such services, large scale data centers are deployed at various geographical locations, and the contents are replicated at multiple locations to achieve better availability and reliability. However, it is not clear given these multiple data centers how content providers place the content and how a user query is served from the cloud. In this paper, we propose a comprehensive, yet simple and intuitive, approach that can be employed to understand the data center driven content distribution using active measurements. Additionally, applying this method on YouTube, we find that YouTube uses data-centers in more than 50 locations, and that it employs an interesting "3-step" approach to deliver videos to clients in a location-aware manner.