Distributed architecture for large-scale video servers

For commercial video-on-demand services, large-scale video servers are needed that can provide 10,000 or more video streams. We propose a distributed architecture for such large-scale video servers. Under the distributed architecture, functionally distributed modules operate independently to retrieve video segments from many disks, and these modules work cooperatively to send continuous video streams to terminals. The problem of balancing loads over modules is overcome by a newly-proposed retrieval method based on retrieval priorities. We also describe a method for finding the optimal allocation pattern to allocate segments in a way that avoids imbalance of the loads in advance. This method converts the segment allocation problem into the form of a constraint optimization problem based on the access probabilities of segments. Solving the optimization problem with a genetic algorithm (GA) leads to the optimal allocation. In basic simulations, our method was able to find an allocation pattern that was superior to other conventional patterns.