Method of predicting number of on-demand video requests using time series data for video cache system

We propose a method of predicting the number of requests for video titles in an on-demand video delivery system considering their time dependency. To handle many heterogeneous video requests in a video streaming delivery network system using video servers, it is effective to introduce video caching systems in the network system. For the cache system design in video delivery network systems, an effective cache algorithm for predicting how many requests will occur in the future is needed. Since almost all such proposed algorithms are for WWW systems, they do not consider the time dependency of the number of requests, so they are not applicable to a video system with time-dependent requests because a customer in the system watches one video title just one time. In our method, the number of requests is represented by a function of time by considering how a customer obtains information about a video and requests it and by considering that a customer requests each video only once. We have developed a method of calculating the coefficients of the function from the series of the number of requests and for estimating the number of requests in the future. We evaluated the convergence and calculation time. The results show that the calculations converge and the calculation time is 31 ms. They also show that the calculation cost is acceptable for a general video delivery system.