Analysis Of Users' Request Pattern And Performance Optimization Of IP Network Video Service System

With the growing of the market demand for internet video service, techniques of this filed have attracted more and more focus. How to optimize the performance of the service system to support higher quality of service with limited resource is a hot spot of research in recent years.Most of the optimizing operation of the internet video service should have to face the problem that users'access request is concentrated and busty. Three points should be considered to resolve this problem:·The factors of service system that lead to the concentration must be clarified.·Build a suitable model to describe the variation trend of the system's load.·Adopt effective policies of load-balancing and call admission control.According to these three points, the work in this dissertation is mainly about the policies and algorithms for optimizing system's performance based on user behavior model, including the user arrival model, the prediction of user arrival, the call admission control policy and the cooperative policy of server clusters based on the user arrival model. The work is as follows:(1)Analyzing the multi-pattern features of users'arrival in internet video service system, using the time serial model to describe the arrival rate, and proposing a novel prediction model based on Markov theory to predict the users'arrival rate.(2)Proposing the fisher optimal division to divide the history data for detecting patterns of users arrival, proposing the GMM to fit the users arrival curve and calculate the joint distribution of the time serial point for improving the accuracy of the Markov prediction model.(3) Analyzing the bottleneck of the video server's resource, and using the queuing theory to build a model of the servers'request condition. Proposing a sub-period call admission control policy based on prediction of the users arrival rate. According to the existing call admission control algorithm can not be used in on-line calculation, a novel algorithm QDTP is proposed to resolve this problem and the PSO is used to accelerate the calculating speed of the algorithm.(4)Analyzing the condition with multiple remote servers that each server has different capacity and has its own user access request pattern. Proposing the cooperative policy to increase the capacity of the whole system, and a novel content routing algorithm based on user arrival prediction is proposed to realize this policy.The method and algorithm proposed in this paper have been proved with the experiment based on the actual data set or simulation.