On-line storage management with distributed decision making for content-centric networks

Abstract

Content traffic proliferation in Internet makes more dire than ever the development of radical new network architectures, where information will be addressed by semantic attributes rather than the origin and destination identities. In this direction, content-centric networking appears as a flexible communication model that meets the requirements of the content distribution trends of the future Internet. In such networks, information will reside at various locations/nodes (the Content Delivery Network surrogate servers) and the requests of the users for some piece of information will be directed to the closest replica. Since the location of the users and the popularity of the content varies over time, the problem of finding the optimal replication pattern for the available content, given the storage constraints, comes into the foreground. In this paper, we propose two on-line storage management algorithms of gradient descent type, designed specifically for content-centric networks. The proposed algorithms are of polynomial complexity and thus adapt easily to any environmental changes. Each node re-assigns its information items with the aim to minimize the overall traffic cost of the content delivery as the popularity and locality of users' requests change. While both the proposed algorithms operate in a distributed way, differ in the amount of information required for the decision making. Thus, we identify the inherent information - performance tradeoff and compare them in terms of network traffic, convergence speed and amount of circulated information.