Interdependent Multi-layer Spatial Temporal-based Caching in Heterogeneous Mobile Edge and Fog Networks

Applications and services hosted in the mobile edge/fog networks today (e.g., augmented reality, self-driving, and various cognitive applications) may suffer from limited network coverage and localized congestion due to dynamic mobility of users and surge of traffic demand. Mobile opportunistic caching at the edges is expected to be an effective solution for bringing content closer and improve the quality of service for mobile users. To fully exploit the edge/fog resources, the most popular contents should be identified and cached. Emerging research has shown significant importance of predicting content traffic patterns related to users' mobility over time and locations which is a complex question and still not well-understood. This paper tackles this challenge by proposing K-order Markov chain-based fully-distributed multi-layer complex analytics and heuristics to predict the future trends of content traffic. More specifically, we propose the multilayer real-time predictive analytics based on historical temporal information (frequency, recency, betweenness) and spatial information (dynamic clustering, similarity, tie-strength) of the contents and the mobility patterns of contents' subscribers. This enables better responsiveness to the rising of newly high popular contents and fading out of older contents over time and locations. We extensively evaluate our proposal against benchmark (TLRU) and competitive protocols (SocialCache, OCPCP, LocationCache) across a range of metrics over two vastly different complex temporal network topologies: random networks and scale-free networks (i.e. real connectivity Infocom traces) and use Foursquare dataset as a realistic content request patterns. We show that our caching framework consistently outperforms the state-of-the-art algorithms in the face of dynamically changing topologies and content workloads as well as dynamic resource availability.