DIS Reference Page

To support large-scale Distributed Interactive Simulation (DIS) exercises effectively, scalable approaches to supporting information flow among the individual distributed simulators must be found. The issue of interest here is whether DIS should be supported by a single homogeneous mode of multicast across all participating local area networks (LAN) and wide area networks (WAN) or by separate multicasting schemes on LAN and WAN.

In conjunction with Dr. Mark Pullen, I investigated an approach called Dual-Mode Multicast that uses multicast at the global (WAN) level on a group-per-exercise basis to minimize WAN traffic, while using multicast at the local (LAN) level to limit the number of PDUs delivered to each simulation. This combines the benefits of multicasting with simplicity at the global (WAN) level of the system and is supportable under existing and projected COTS multicast WAN technologies. To evaluate this approach, we have constructed various scenarios and determined PDU rates in various parts of the system for these scenarios. These scenarios have given us insights into the computational demands of this technique, as well as previously described multicast approaches.

We constructed simulations to study the network level characteristics of this approach on both small (500 entities) and large (50,000 entities) simulations. These simulations provided insight into the computational and communications demands of this technique, including the rate at which incoming PDUs are filtered at the application gateways, the PDU traffic expected on the individual LANs, and the effect of the grid size used for the multicast groupings. We also built simulations using data (terrain, entity distributions and physical distributions) from a real exercise, STOW-97, to get a more accurate picture of what happens in an exercise that uses this techniques.

The emergence of command forces (CFORs) as a simulated entity within a DIS simulation requires effective interfaces between existing learning systems and simulation applications. Existing methods utilize programmers and knowledge engineers to encode the knowledge of a subject matter expert (SME) into a knowledge base. However, there are no interfaces that integrate existing learning systems with a DIS application that would provide effective and efficient knowledge acquisition by allowing the SME to directly instruct the learning system.

We integrated Disciple, a machine learning system, with ModSAF to provide this direct interface for the SME. This integration extends the familiar graphical interface of ModSAF to include the tools necessary for the SME to instruct an agent through demonstration. A new experimental PDU was developed to provide the communication mechanism and common data format for the two applications to communicate.

Associated Publications