In this paper we describe research and development on the concept and application of distributed terrain and distributed terrain servers to support live/virtual training operations. This includes design of a distributed, cluster-capable “Combat Server” for the virtual representation and simulation of live training exercises, and current work to support virtual representation and visualization of live indoor operations involving firefighters, SWAT teams and/or special operations forces. The Combat Server concept under development is for an object-oriented, efficient and flexible distributed platform designed for simulation and training. It can operate on any compatible, high performance computer for which the software is compliant; however, it is explicitly designed for distribution and cooperation of relatively inexpensive clustered computers, together playing the role of a large independent system. The design of the Combat Server aims to be generic and encompass any situation that involves monitoring, tracking, assessment, visualization and, eventually, simulated interactivity to compliment real-world training exercises. To accomplish such genericity, the design must incorporate techniques such as layering or abstraction to remove any dependencies on specific hardware, such as weapons, that are to eventually be employed by the system; this also includes entity tracking hardware interfaces, whether by GPS or Ultra-Wide Band technologies. The Combat Server is a framework. Its design is a foothold for building a specialized distributed system for modeling a particular style of exercise. The combat server can also be a software development framework, providing a platform for building specialized exercises while abstracting the developer from the minutia of building a real-time distributed system. In this paper we review preliminary experiments regarding basic line-of-sight (LOS) functions of the combat server functionality and scalability in a cluster computing environment. Our initial results show that computing LOS between entities on a distributed terrain scales well over a large number of processors.