GEM-Tree: Tree-Based Analytic Geometrical Multi-Dimensional Content-Based Event Matching

In large content-based multi-attribute publish /subscribe systems, event matching is a key component, which is in charge of finding all subscriptions that match events. However, the increasing user scale, the enriching message diversity, and the aggravating QoS demands make the performance of event matching severely challenged. Most existing event matching schemes cannot efficiently sustain in these scenarios. The performance rapidly drops as the system load rises. In this paper, we propose GEM-Tree (Geometrical Event Matching Tree), a novel tree-based analytic geometrical index structure for highly efficient event matching in large-scale content-based publish/subscribe systems. To further improve the event matching speed, a local-adjustment mechanism is designed to determine the deployment for each new subscription registering into the GEM-Tree, and a global-adjustment mechanism is designed to optimize the locations of the subscriptions already inserted in GEM-Tree. The experiment results in 8 scenarios demonstrate that GEM-Tree is superior to 3 state-of-the-art reference schemes(BE-Tree, OP-Index, and TAMA). Especially, the leading advantage of GEM-Tree is more significant in matching time for a large number of subscriptions.