Dynamic Caching Strategy for Multi-bitrate Videos Based on Popularity Prediction

Abstract

With the rapid increase of video traffic in the Internet, Content Delivery Network (CDN) has been used more widely, which is considered as a conventional method to support video delivery services. Scalable video coding and edge computing provide CDN with the ability to transcode videos at the edge of the network, enabling more flexible delivery of multi-bitrate videos. However, choosing which contents from the video library to cache on the edge servers remains a challenging problem, especially when the problem is extended to dynamic scenario. In this paper, an online caching model for dynamic scenario is designed innovatively, where the popularity of videos and users' requests change over time, while the caching strategy will be adjusted accordingly. Then a two-levels video popularity prediction algorithm (global popularity and local popularity) is novelly proposed, which is adapt to deal with in the caching strategy problem. Based on the popularity prediction, a revenue calculation algorithm is further proposed to obtain the predicted caching revenue of multi-bitrate videos. Finally, we develop a hybrid greedy based genetic algorithm to optimize the caching model, and perform simulation experiments using the YouTube dataset as a simulated dataset. The experimental results show that our popularity prediction based dynamic caching algorithm (PPDA) can greatly improve the cache hit ratio and reduce the average transmission delay.