User Behavior Analysis And System Optimization For CDN Based Video-on-Demand System

The Video on Demand (VoD) service has a long history. With the advances of the network transmission capacity in recent years, Internet-based video-on-demand services rise gradually and become the mainstream applications on the Internet. For example, YouTube is one of the hallmarks of VoD services which have had a tremendous influence. VoD service is characterized by large, high bandwidth. So it is a resource consumption-based service, which could easily lead to high operational costs. User behavior analysis for VoD services is to understand the user behavior characteristics and rules, thereby guiding system design to improve the user experience and reduce operation and maintenance costs. All content providers, network operators and end customers can benefit from it.The main indicators on existing user behavior for video-on-demand services include: video length, video popularity, user access time distribution, the user viewing duration.This paper is based on the VoD services for 2008 Beijing Olympic Games. In this paper, user behavior conducted in-depth analysis and summary. We give a comparative analysis with the traditional film-based VoD service and user-generated content VoD service. By comparing the analysis, we found that the Beijing Olympic Games VoD services show the following unique user behavior rules: 1) the user has a random access time; 2) The user average viewing time is short; 3) 80% of users to watch videos rarely used jump, and pause functions; 4) user access to certain content easily lead to sudden concentration of large numbers of users to access (Flash Crowd) phenomenon.Based on the above analysis for user behavior, this paper presents system design improvements for the different VoD services. The main suggestions for improvement mainly related to the following four points: 1) system structure; 2) content distribution strategy design; 3) content distribution and log analysis system design; 4) load balancing recommendations. The last part of paper is the summary and generalization of the full text, and here I proposed questions for further research and exploration direction.