Design and Implementation of a Fully Distributed Caching Algorithm on an NDN System

Abstract

ICN (Information Centric Networking) is a new method of storing and accessing data on the internet which focuses on the content itself rather than the IP (Internet Protocol) address where the content is stored. ICN enables both in-network caching and name-based data retrieval. This allows for better usage of edge cloud resources, giving the user a faster response time as some data requests and services may be handled locally. NDN (Named Data Networking) is a specific type of ICN which locates and delivers content based on the associated data name rather than using the source or destination host addresses. For NDN to be most beneficial, we need to implement efficient caching algorithms that consider the needs of many users in a network. To address this need, we have developed a caching algorithm for an NDN network in a tree topology. It is fully distributed and makes storage and eviction decisions at each router based on the number of hops needed to retrieve the data and the popularity of the data at that router. The total number of hops taken by all data during the testing period determined the algorithm’s true cost. We tested our algorithm using an NDN testbed and compared its true cost with another commonly used algorithm, LRU (Least Recently Used), under the same conditions. Our cost-based policy incurred a lower true cost in all test cases, with average savings ranging from 9% to 19% depending on cache size and popularity distribution. The cost policy performed particularly well in comparison to LRU when the cache size was small.