Adaptive Edge Caching and Routing in Named Data Networking of Things Using Contextual Bandits and SDN Controllers

Abstract

Named Data Networking is a promising architecture for the Internet of Things, since its in-network caching can hold content close to consumers and shorten retrieval delay. In most caching policies, however, the decision is made locally at each router, which is often unaware of changing network conditions. This paper presents Adaptive Edge Cache (AEC), a scheme that combines in-network caching, software-defined control, and online learning. AEC divides the network into clusters, each managed by a controller that maintains a shared table assigning every cluster router the content hashes it must cache. Controllers monitor router health through lightweight periodic messages, and whenever the set of working routers changes, a learning agent inside the controller reassigns the hashes to new caching homes and updates the forwarding state so that requests avoid unavailable or depleted routers. AEC and five representative baselines were implemented in ndnSIM and evaluated on a 131-node topology, sweeping cache size over 5, 10, 50, and 100 objects, popularity skew over 1.2, 1.6, and 2.0, and the fraction of failed routers over 0, 30, 40, and 50 percent. AEC gains most at small cache sizes and low popularity skew, while keeping delay, hop count, and eviction overhead moderate. Under router failures it adapts its caching homes and forwarding state, substantially reducing delay and retransmissions and avoiding the severe degradation seen with static caching and forwarding.