24 April 2014
De Vere Wokefield Park
Europe/London timezone

CD-GAIN: Measuring Traffic Gains from Peer-assisted Content Delivery of Long Duration Video-on-Demand content

24 Apr 2014, 17:00
30m
De Vere Wokefield Park

De Vere Wokefield Park

Goodboys Lane Mortimer Reading Berkshire RG7 3AH
Main Session

Speaker

Dr Dmytro Karamshuk (King's College London)

Description

Efficient delivery of video content is a critical problem. Video traffic now accounts for 57% of all consumer traffic on the Internet and is forecasted to reach 69% by 2017. To handle these massive volumes of traffic content delivery networks (CDNs) are turning to clients for assistance, creating hybrid peer-assisted content delivery systems. In practice, peer-assisted distribution of video-on-demand content faces a number of design obstacles which include: the need of localizing peer-to-peer traffic within ISPs (isp-friendliness), reluctance of users to participate in redistributing the content (partial participation) and necessity to match users with similar bitrate requirements (bitrate stratification).

Summary

In this paper we study these problems by analyzing a month-long trace of 15M accesses to BBC iPlayer in London. Our results show that up to 80% of server traffic can be saved by employing peer-assistance even despite the obstacles. Particularly, ISP-friendliness has insignificant impact on the traffic savings as the majority of user accesses come from a small number of large ISPs. Similarly, most of users watch content in one of two most popular bitrate formats. On the other hand, partial participation can significantly reduce the benefits of peer-assisted content distribution. To drive intuition, we develop a simple analytical model which links traffic gains from peer-assistance to the average number of users in peer-to-peer swarms, i.e., capacity of the swarms, and show how various obstacles impact the capacity. We also investigate the impact of two well-known techniques for improving content availability on the traffic gains and show that bundling is not effective in this context, however, a simple caching approach can boost the gain from peer-assistance for up to 30%.

Primary author

Dr Dmytro Karamshuk (King's College London)

Co-author

Dr Nishanth Sastry (King's College London)

Presentation materials