On the feasibility of prefetching and caching for online TV services: a measurement study on hulu

  • Authors:
  • Dilip Kumar Krishnappa;Samamon Khemmarat;Lixin Gao;Michael Zink

  • Affiliations:
  • University of Massachusetts Amherst;University of Massachusetts Amherst;University of Massachusetts Amherst;University of Massachusetts Amherst

  • Venue:
  • PAM'11 Proceedings of the 12th international conference on Passive and active measurement
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Lately researchers are looking at ways to reduce the delay on video playback through mechanisms like prefetching and caching for Videoon-Demand (VoD) services. The usage of prefetching and caching also has the potential to reduce the amount of network bandwidth usage, as most popular requests are served from a local cache rather than the server containing the original content. In this paper, we investigate the advantages of having such a prefetching and caching scheme for a free hosting service of professionally created video (movies and TV shows) named "hulu". We look into the advantages of using a prefetching scheme where the most popular videos of the week, as provided by the hulu website, are prefetched and compare this approach with a conventional LRU caching scheme with limited storage space and a combined scheme of prefetching and caching. Results from our measurement and analysis shows that employing a basic caching scheme at the proxy yields a hit ratio of up to 77.69%, but requires storage of about 236GB. Further analysis shows that a prefetching scheme where the top-100 popular videos of the week are downloaded to the proxy yields a hit ratio of 44% with a storage requirement of 10GB. A LRU caching scheme with a storage limitation of 20GB can achieve a hit ratio of 55% but downloads 4713 videos to achieve such high hit ratio compared to 100 videos in prefetching scheme, whereas a scheme with both prefetching and caching with the same storage yields a hit ratio of 59% with download requirement of 4439 videos. We find that employing a scheme of prefetching along with caching with trade-off on the storage will yield a better hit ratio and bandwidth saving than individual caching or prefetching schemes.