While distributed simulation infrastructures have evolved dramatically over the past several years to provide ever increasing levels of flexibility, abstraction, and interoperability, the scalability and performance of the simulation infrastructure continues to be a critical limiting factor. In particular, it is now becoming apparent that the limitations of the supporting networking technologies are a significant impediment to achieving needed levels of scalability and performance. Advancing the state-of-the-art for large-scale distributed simulations therefore requires significant advances both in the underlying network technologies and in the ability of simulations to exploit these new capabilities.
Under the Specialized Active Networking technologies for Distributed Simulation (SANDS) project sponsored by the Information Technology Office (ITO) of the Defense Advanced Research Projects Agency (DARPA), TASC and the University of Massachusetts, Amherst (UMass) are developing Active Networks-based capabilities to improve significantly the performance of network-based distributed simulations1. Our primary objective is reducing the substantial amounts of irrelevant network traffic delivered to simulation hosts in order to both improve bandwidth efficiency and to reduce the considerable overhead associated with reading and discarding unneeded data. Our approach involves installing dynamic packet filters within the network that act on behalf of each host to eliminate unneeded packets as early as possible. Our goal is a seamless integration with the High Level Architecture (HLA) Declaration Management (DM) and Data Distribution Management (DDM) services.
Use of Active Networks to provide interest management services offers several important benefits to large scale simulations: (i) Because each entity can install its own filters, information filtering is accomplished in a "receiver-driven" manner, allowing each entity to customize its filters according to its own need. This decentralized approach allows active filtering to scale well as the number of entities grows large. (ii) Because active filtering is performed at a routing point, filtering can also be dependent on the state (e.g., congestion-level) at that router. In particular, this allows both entities and network routers to determine which data should be shed in times of congestion overload, and provides an effective means for mediating among the conflicting demands of different entities.