Contextual Multi-armed Bandits for Data Caching in Intermittently-Connected Lossy Maritime Networks

Information centric networking (ICN) is a promising architecture for intermittently-connected lossy maritime networks. By decoupling content from location and storing popular content in caches ICNs enable content access even when content producers are disconnected from the network. However, network-resource constraints, node mobility, and low data rates remain as major challenges to achieve the potential benefits that increased data bring to maritime networks. In this work, we propose an ICN architecture that uses dedicated caches and off-path caching to deliver popular network content to consumers. Our architecture motivates a data-caching problem in which each cache must define effective and efficient caching policies, with effectiveness assessed based on the cache-hit ratio achieved by the cache and efficiency assessed based on the usage of the communications budget allocated for data caching purposes. This work develops an online caching algorithm that enables dedicated caches to sequentially decide what content to cache. The online caching decision making problem is formulated using the contextual multi-armed bandits framework. The performance of the online caching algorithm is illustrated via simulation examples on synthetic data.