Section: New Results

Cache Management in CCN

Participants : Thomas Silverston [contact] , César Bernardini, Olivier Festor.

The Internet is currently mostly used for accessing content. Indeed, ranging from P2P file sharing to current video streaming services such as Youtube, it is expected that content will count for approximately 86% of the global consumer traffic by 2016.

While the Internet was designed for -and still focuses on- host-to-host communication (IP), users are only interested in actual content rather than source location. Hence, new Information-Centric Networking architectures (ICN) such as CCN, NetInf, Pursuit have been proposed giving high priority to efficient content distribution at large scale. Among all these new architectures, Content Centric Networking (CCN) has attracted considerable attention from the research community (http://www.ccnx.org ).

CCN is a network architecture based on named data where a packet address names content, not location. The notion of host as defined into IP does not exist anymore. In CCN, the content is not retrieved from a dedicated server, as it is the case for the current Internet. The premise is that content delivery can be enhanced by including per-node-caching as content traverses the network. Content is therefore replicated and located at different points of the network, increasing availability for incoming requests.

As content is cached along the path, it is crucial to investigate the caching strategy for CCN Networks and to propose new schemes adapted to CCN. We therefore designed Most Popular Content (MPC), a new caching strategy for CCN network [10] .

Instead of storing all the content at every nodes on the path, MPC strategy caches only popular content. With MPC, each nodes count all the requests for a content and when it has been requested a large amount of time, the content will be cached at each node along the path. Otherwise, the content is not popular; it is transmitted but it is not cached into the network.

We implemented MPC into the ccnSim simulator and evaluate it through extensive simulations.

Our results demonstrate that using MPC strategy allow to achieve a higher Cache Hit in CCN networks and still reduces drastically the number of replicas. By caching only popular content, MPC helps at reducing the cache load at each node and the network resource consumption.

We expect that our strategy could serve as a base for studying name-based routing protocols. Being a suggestion based mechanism, it is feasible to adapt it to manage content among nodes, to predict popularity and to route content to destination. In addition, we are currently investigating the social relationship between users to improve our caching strategy for CCN networks.