A Distributed Whiteboard Infrastructure for Information Fusion Simulations

摘要

We argue that architectural ideas from DeeDS, a distributed real-time database, can be used to create a whiteboard infrastructure usable in information fusion applications as well as in fault tolerant simulations. We discuss the need for fault-tolerant real-time simulation in information fusion, and how this can be supported by the whiteboard infrastructure. There are many reasons to perform real-time simulations (including environment simulation) - the use of real hardware or systems can be prohibitively expensive or dangerous to people and equipment, or may simply not be available at all. The combination of real and simulated objects, or nodes, is of particular interest when it is prohibitively expensive, dangerous, or impossible to test sufficient numbers of collaborating nodes in a realistic application. The need for fault tolerance of such real-time simulations, with a mixture of real and simulated nodes, occurs because it is difficult/impossible to restart physical nodes or wait for restart of simulated nodes, if one or more simulation instances fail. The key problem when mixing real and simulated nodes is the occurrence of "external actions" that cannot easily be undone and redone when a failure has occurred. We describe a natural approach to support multiple degrees of fault tolerance in real-time simulations, based on optimistic synchronization on top of a whiteboard architecture. The whiteboard architecture is a natural and useful infrastructure metaphor for information fusion, an information-based exchange that allows real and simulated nodes to be freely mixed.