Optimal Caching for H.264 Partitioned Video Streaming

In this paper we investigate the possible benefits of a caching scheme for video encoded using the data partitioning option from the H.264 standard. We propose a method to measure the importance of a partition based on its contribution to the peak signal-to-noise ratio (PSNR) on all affected frames, and we term this measurement as PSNR profit. We calculate the average PSNR profit for each frame and partition type of real H.264 video footage. We then apply the Knapsack model to the storage optimisation of the partitioned video in terms of the total profit on a caching server. We show that the optimized cache (OC) outperforms the conventional cache (CC) when the cache reaches its maximum capacity and new videos continue to arrive. Under these circumstances, optimal caching aims to keep as many the most important partitions as possible, while conventional caching treats all the partitions equally. As a consequence, the server profit of an OC will continue to rise as the number of videos in the system increases, compared to the unchanged profit of a CC; and the client profit (relating to the received video quality) will reduce at a slower rate than that of the CC option. We also show that optimal caching can provide benefit even in a lossy environment because in this scenario the client profit is degraded more gracefully than that in the un-optimized scheme. When the optimized caches are formed as a caching network, it can be ensured that all the partitions of each video will remain somewhere in the network, instead of being deleted entirely as happened in the case where a sole cache is used. The required number of the servers or the proportion of space allocated for long term storage of videos in the network can then be determined.