Patrick Tendick

Publish-subscribe services for urgent and emergency response

In urgent and emergency response situations, publish-subscribe services need to go beyond information dissemination to facilitate response collection, and even collaboration, among the recipients. We introduce flexible delivery, role-based subscription guidance, and historical event matching to address the requirements of urgent response applications. Flexible delivery allows publishers to choose the most appropriate communication technique for the urgent situation. Role-based guidance provides an interface for subscribing to events from the users perspective. Historical event matching allows subscribers to join ongoing collaborations for events that occurred in the past. Together theses techniques allow the creation and support of ad hoc communities of interest to address urgent situations. We report our experience with these techniques during 3 years of production use for escalating product repair issues for our company. Forum provides the first production use cases that require historical matching of persistent events in publish-subscribe services.

Decisions as a service for application centric real time analytics

The need for application-level intelligence cannot be easily satisfied with existing architectures or methodologies that separate methods and tools for application developers and data scientists. We aim, therefore, to develop a framework (an architecture and a methodology) to make it possible to add intelligence capabilities to existing applications (decision-enablement) and to facilitate building new decision-enabled applications. The proposed approach starts by instrumenting the existing code with logging and instrumented decision points. The execution of the application produces information that is initially used to reengineer its behavior and then the decision points are used to conduct search-based experimentation to optimize its behavior. This lightweight instrumentation allows the application developer and data scientist to fully exploit their capabilities, with the framework providing the glue needed to put their work together easily and transparently. In particular, the analytic capabilities, such as analysis of operational data, are better dealt with in the decision-making part, without complicating the mechanics of how the application functions. We plan to apply this framework to decision enable existing systems and to build new systems from scratch and measure the effectiveness of the approach and of the resulting products.