Content distribution for publish/subscribe services

Caching and content delivery are important for content-intensive publish/subscribe applications. This paper proposes several content distribution approaches that combine match-based pushing and access-based caching, based on users' subscription information and access patterns. To study the performance of the proposed approaches, we built a simulator and developed a workload to mimic the content and access dynamics of a busy news site. Using a purely access-based caching approach as the baseline, our best approaches yield over 50% and 130% relative gains for two request traces in terms of the hit ratio in local caches, while keeping the traffic overhead comparable. Even when the subscription information is assumed not to reflect users' accesses perfectly, our best approaches still have about 40% and 90% relative improvement for the two traces. To our knowledge, this work is the first effort to investigate content distribution under the publish/subscribe paradigm.