Pruning subscriptions in distributed publish/subscribe systems

Publish/subscribe systems utilize filter algorithms to determine all subscriptions matching incoming event messages. To distribute such services, subscriptions are forwarded to several filter components. This approach allows for an application of routing algorithms that selectively forward event messages to only a subset of filter components. Beneficial effects of this scheme include decreasing network and computational load in single filter components. So far, we can find routing optimizations that exploit coverings among subscriptions or utilize subscription merging strategies. Generally, such optimizations aim at reducing the amount of subscriptions forwarded to filter components, which decreases their computational load. This might in turn result in an increasing number of event messages routed through the network. However, current optimization strategies only work on restrictive conjunctive subscriptions and cannot be extended to efficiently support arbitrary subscriptions. Furthermore, it is not possible to apply covering and perfect merging strategies in all application scenarios due to the strong dependency of these approaches on actually registered subscriptions. In this paper, we present a novel optimization approach, subscription generalization, to decrease the filtering overhead in publish/subscribe systems. Our approach is based on selectivities of subscriptions and can be utilized for all kinds of subscriptions including arbitrary Boolean and conjunctive subscriptions. We propose a simple subscription generalization algorithm and show an evaluation of the results of a first series of experiments proving the usefulness of our approach.