Department of Computer Science & Engineering

Content-based publish / subscribe (CBPS) systems are a natural substrate for context-aware applications because they provide the right separation of concerns, efficient event distribution, extensibility, and scalability. However, the separation of concerns afforded by CBPS middleware precludes publishers from collaborating with each other to efficiently detect context-aware conditions for publication as events. We overcome this problem with an open implementation approach, which enables subscribers to attach a domain-specific implementation strategy to their context-aware subscriptions. Strategies are supported by first-class active subscriptions that can deploy lower-order dynamic publisher / publisher cross-subscriptions where events enter the CBPS system. By exchanging data on an as-needed basis, event traffic can be dramatically reduced. Our algorithm for detecting the proximity of mobile buddies reduces event traffic from O( |events| ) to O( |movement| ) worst case, O( lg2|movement| ) expected. Experiments reveal an 82% reduction with current positioning technologies.

Pre-2018 CSE ID: CS2004-0791