Efficient document filtering and routing methods in publish/subscribe systems

Abstract

Publish/subscribe systems are frequently used for disseminating documents to various clients geographically distributed in the network. Typically, such systems consist of publishers who produce documents, subscribers who consume them, and a broker that delivers documents from publishers to subscribers. If the publish/subscribe system maintains multiple brokers to achieve scalability, a document routing method between brokers is needed. To do this, in the first part of this dissertation, we propose an efficient subscription routing algorithm with subscriptions especially expressed in XPath patterns as in an XML-based publish/subscribe system [55]. We first discuss that publish/subscribe systems can utilize the homomorphism relationship among XPath patterns to route XPath patterns. Then, we develop a lattice data structure called the partially ordered set of XPath patterns (abbreviated as POX) and its corresponding algorithm for efficiently maintaining homomorphism relationships. Finally, we present performance evaluation results that validate our algorithm with respect to system performance and scalability. In the second part of this dissertation, we mention the $\textsl{k}$ -NN query that the publish/subscribe system can frequently encounter and develop algorithms for exact $\textsl{k}$ -NN search using the R-tree that is the most popular structure to index multi-dimensional data items. Our method pre-computes and maintains l nearest neighbors for each subscription and exploits them to prune the search space. To minimize the cost of reading the l-NN subscriptions from the disk, we also propose a method that expects l’, the minimum value of l, according to the value $\textsl{k}$. We prove that our method always returns exact results and propose complete algorithms that process the $\textsl{k}$ -NN query and update the R-tree. The experimental results show that our method outperforms the existing one proposed in [49]. In Mobile Ad Hoc Networks (MANETs) wher...