Buffer management in probabilistic peer-to-peer communication protocols

摘要

In multipeer communication decentralized probabilistic protocols have received a lot of attention because of their robustness against faults in the communication traffic and their potential to provide scalability for large groups. These protocols provide a probabilistic guarantee for a propagated event to reach every group member. Recent work aims to improve the scalability of such protocols by reducing memory requirements. In saving memory resources, the history buffer, which is used to "remember" received events and to prevent multiple deliveries of events to the application, plays a very significant role. We examine how the buffer size should be chosen to challenge the multiple delivery problems. Further, we propose and evaluate several methods of optimizing the dissemination of events in order to provide high reliability and reduce the number of multiple deliveries at the same time.