Two- Time-Scale Hybrid Proactive and Reactive Edge Caching for Content Delivery Networks

Abstract

In content delivery networks (CDNs), there usually exist two types of storage disks named hard disk drive (HDD) and solid state disk (SSD), which have different read speeds and cache sizes. By accounting such a practical hardware structure of CDNs and time-variant content popularity, we propose a novel two- time-scale caching strategy to determine the content placement in HDDs and SSDs. Specifically, we first propose a Stratification and Clustering based Ridge regression model with Laplacian regularization (SCRL) to predict the dynamic content popularity by utilizing both the video feature and user preference. The additional Laplacian regularization improves the prediction performance by leveraging the neighboring information of each video. Based on the predicted results, we then design an efficient hybrid proactive and reactive cache replacement strategy (HPRR) on a two-time-scale basis. Real-world trace-based numerical results show that the proposed prediction and caching strategy can significantly outperform the considered existing methods.