Popularity estimation based video caching mechanism for VoD service

Abstract

Nowadays, people are now using more web connected devices than ever, so the people uses web services whenever and wherever. They check e-mail, watch videos, or surfing web pages at bus, train, or street. Moreover, as the devices` specs is getting higher, people wants much higher quality of service, and it causes excessive bandwidth traffic. Video traffic is the one of the highest traffic of the overall network traffic. Watching movie, drama, news, or web pages requires video transmission, so that those service needs considerable network bandwidth. In that sense, reducing video traffic is very important, and it is also related to the overall network traffic. Caching video contents in proxy, or gateway server is one of the most effective, and practical solution for reducing network traffic. The proxy server stores several videos and serves clients` request. If a client request a video, and a proxy server has the video, then the clients can get the video from the proxy server, not a origin video server. It implies that the network delay decreases, and the overall network bandwidth use diminishes; which gives quite higher quality of service for clients. In this paper, I proposed the popularity estimation based video cache mechanism. For the effective video caching, we need to know which video will be watched in future. Storing more popular video, rather than less popular video, can serves more clients` request, then the cache utilization gets higher. Therefore, estimating the future popularity for the videos is very important. The popularity of videos can be affected by many factors. Some event may cause sudden change of the video popularity. The conventional video popularity estimation, however, is based on simple popularity estimation, so it cannot catch the sudden change of popularity. For this reason, new factor, the number of shared, is added to the conventional estimation method, and it can catch the sudden change. With real data from YouTube, we simulated s...