An Operator Approach to Popularity-Based Caching in DHTs

Caching is a long investigated topic in the context of P2P networks and distributed hash tables (DHTs). Most of the proposed caching algorithms are user-centric in the sense that they focus on minimizing the search size or latency as the target metrics. In contrast, we take the network operator-centric view in this paper. We set the total traffic generated by the DHT search as the metric of interest and provide a simple algorithm to achieve a maximum reduction of the traffic. The key of the algorithm is that every node maintains statistics about routed queries and replicates its cached objects only when the popularity of queries for those objects are above a certain threshold. Through extensive simulations, we determine the optimal value of the popularity threshold and show the traffic reduction in comparison with state of the art DHT caching solutions. Our algorithm achieves an average traffic reduction of around 30% as compared to caching along the entire search paths, while maintaining good performance in terms of user-oriented measures such as the number of search hops, or equivalently, the search latency.