syslog
28Mar/110

Tracking shared YouTube links might help improve video delivery performance

Even though sharing YouTube links with our friends on Twitter and Facebook might seem a simple and quick action with no consequences, when millions of users are doing so every day it is something much more important. In fact, it becomes possible to track how interest for content items is spreading across social networks to improve performance of the content delivery networks serving those items. This is the problem we have addressed in a recent paper which will appear at the upcoming WWW 2011 conference.


Online video consumption is soaring, and the steady and impressive growth of YouTube epitomizes this trend. However, the success of sites such as YouTube is not only arising from the entertainment provided to final users but also form the focal point of social interaction that revolves around content. Users prefer consuming online videos also because it offers the opportunity to interact and chat about content, as well as to share it with simple acts as sending a link to a friend. As a result, content diffusion is fostered by these links shared on online social networks posts, which may often generate floods of requests to the provider through a social cascade. For instance, on Twitter there are more than 400 tweets per minute with a YouTube link and a video shared on Twitter spawns on average 6 more viewing sessions.

Our proposal takes advantage of the fact that social cascades can propagate in a geographically limited area to discern whether an item is spreading locally or globally. This informs cache replacement policies, which utilize this information to ensure that content relevant to a cascade is kept close to the users who may be interested in it. Our analysis of a novel Twitter dataset shows how social cascades are likely to spread on geographically local distances. Users tend to share content over short-distance social connections, despite the presence of several long-range links: about 40% of steps in social cascades involve users that are, on average, less than 1,000 km away from each others.

The main result of this work is that geographic information about social interactions can be extracted and used to improve large-scale system design, as content consumption becomes more and more social. We see a great potential in exploiting geographic properties of human communication on online services. Geographic locality of online interactions can be exploited to do pre-fetching of Web content, caching of normal HTTP traffic, distributed database solutions for online social platforms and even to devise security mechanisms.

Track Globally, Deliver Locally: Improving Content Delivery Networks by Tracking Geographic Social Cascades - Salvatore Scellato, Cecilia Mascolo, Mirco Musolesi, Jon Crowcroft - In Proceedings of 20th International World Wide Web Conference (WWW 2011). Hyderabad, India. March 2011. [PDF]