Topology-Aware Cooperative Caching in an Online Learning Environment

Abstract

Previous research has excelled in designing caching algorithms to respond to different arrival processes of requests to the caching nodes, but there remains a considerable gap in the literature with regards to how the topological features of a network affect the performance of a caching algorithm. This thesis presents a framework for incorporating topological information such as betweenness centrality and inter-node distances into the caching decisions made at each node on the network. Through simulations under different item popularity distributions and different network topologies, I conclude that the algorithm presented in the thesis can outperform a topologically-oblivious LFU-based selfish caching strategy and achieve lower network-wide delay when serving requests. This is most apparent in cases where nodes have similar popularity distributions over the library of items, and the network is primarily one connected component.