Scalable time management algorithms using active networks for distributed simulation

This paper investigates time management for distributed simulations in active networks. Time management is essentially the computation of the Lower Bound Time Stamp (LBTS) across federates in a distributed simulation, including in-transit (in-flight) messages. We show that the LBTS computation is an instance of the Distributed Termination Detection (DTD) problem and how DTD algorithms can be applied to LBTS in an active network. These algorithms are potentially much more efficient than traditional LBTS algorithms that rely on point-to-point communication. We introduce the Distinguished Root Node algorithm that can compute LBTS in O(log n) time in general network topologies using a prototype implemented in Java. Experimental results are reported for the Active Time Management Daemon prototype implementation in (1) a simple testbed configuration, (2) a larger cluster environment, (3) an ns-2 simulation of very large configurations, and (4) integrated into an actual HLA-compliant run-time and simulation. While active network algorithms have the potential for improved functionality and superior performance, this potential will only be fully realized when "native" implementations are possible. For wide-scale deployment, real-world issues such as reliable delivery dynamic routing topologies, security and fault tolerance will have to be systematically addressed Also, in the area of distributed computing known as grid computing, the management of distributed resources is a key issue. Hence, we plan to use the grid computing infrastructure to manage overlays and facilitate the use of active networks by large-scale, distributed grid applications.