Estimation of Worst-Case Bandwidth Requirements of Video-on-Demand Systems with Replica Servers using the M/G/¥ Model

This study assesses the worst-case demand and replication models for determining the bandwidth required by the origin server when a number of video programs in a videoon-demand system are replicated on replica servers. The bandwidth required by the video-on-demand system determines the number of viewers that could watch video programs simultaneously. If a video program is viewed in its entirety, the required bandwidth is estimated based on the number of concurrent viewers as well as the length of the video program. Unlike traditional research models that use the hit ratio to assess the replication benefits, this study uses the M/G/ model to examine the related phenomenon and problems in this field. We develop a selection model for replication using the product of the viewing request probability and the video length as the main factor in determining the videos selected for replication. This model theoretically outperforms traditional models that consider only the viewing request probability in the benefits of reducing the worst-case demand at origin server, and the simulation results validate the theoretical results. Due to resource constraints in the replica server, discussions and verifications on the optimization of the proposed replication model were carried out and the findings were used as the basis for the development of a heuristic algorithm. The results derived using this heuristic algorithm are almost the same as the optimized results derived from LINGO.