Storage planning and replica assignment in content-centric publish/subscribe networks

摘要

Content-centric publish/subscribe networking is a flexible communication model that meets the requirements of the content distribution in the Internet, where information needs to be addressed by semantic attributes rather than origin and destination identities. In current implementations of publish/subscribe networks, messages are not stored and only active subscribers receive published messages. However, in a dynamic scenario, where users join the network at various instances, a user may be interested in content published before its subscription time. In this paper, we introduce a mechanism that enables storing in such networks, while maintaining the main principle of loose-coupled and asynchronous communication. Furthermore, we propose a new storage placement and replica assignment algorithm which differentiates classes of content based on their popularity and minimizes the clients response latency and the overall traffic of the network. We also present and compare two replica assignment alternatives and examine their performance when both the locality and the popularity of users request change. The performance of our proposed placement and replica assignment algorithm and the proposed storing mechanism is evaluated via simulations and insights are given for future work. The proposed mechanism is compared with mechanisms from the CDN (Content Delivery Networks) context and performs as close as 1-15% (depending on the conducted experiment) to a greedy (near optimal) approach installing up to 3 times less storage servers in the network and providing the necessary differentiation among the classes of the content.