A basic model for proactive event-driven computing

Abstract

During the movie "Source Code" there is a shift in the plot; from (initially) reacting to a train explosion that already occurred and trying to eliminate further explosions, to (later) changing the reality to avoid the original train explosion. Whereas changing the history after events have happened is still within the science fiction domain, changing the reality to avoid events that have not happened yet is, in many cases, feasible, and may yield significant benefits. We use the term proactive behavior to designate the change of what will be reality in the future. In particular, we focus on proactive event-driven computing: the use of event-driven systems to predict future events and react to them before they occur. In this paper we start our investigation of this large area by constructing a model and end-to-end implementation of a restricted subset of basic proactive applications that is trying to eliminate a single forecasted event, selecting between a finite and relatively small set of feasible actions, known at design time, based on quantified cost functions over time. After laying out the model, we describe the extensions required of the conceptual architecture of event processing to support such applications: supporting proactive agents as part of the model, supporting the derivation of forecasted events, and supporting various aspects of uncertainty; next, we show a decision algorithm that selects among the alternatives. We demonstrate the approach by implementing an example of a basic proactive application in the area of condition based maintenance, and showing experimental results.