The Marine Corps' Live, Virtual and Constructive Training Environment (LVC-TE) connects training systems at geographically separate bases to enable collective and battle staff training. The long-haul circuits that provide the connections are not dedicated to training exercises but are shared and simultaneously carry other network traffic for the Marine Corps. Excess latency and jitter injected into training exercises from these circuits can invalidate results and bias the results of the exercise for one side. A major existing deterrent to the planning of large scale exercises is the inability to accurately estimate the load that will be placed by a local, regional, or country-wide training exercise on the underlying communication networks. This significantly prolongs the planning and approval processes. In this paper, we present a new simulation-based framework to predict the impact of connecting training systems across different types of long-haul network circuits, validate key performance parameters, and streamline the planning of distributed training exercises. The framework profiles different training simulations/simulators and correlates captured traffic to scenario events. Traffic models can be scaled to represent higher numbers of entities, simulators, and time-varying, overlapping scenario events. Authoritative Marine Corps descriptions of the network on which the training exercise is run, in the form of Visio or similar formats, are converted into an executable, dynamic network simulation model. The traffic models are overlaid on the simulated network to predict how traffic generated during a training exercise, competing with non-training traffic, will be delivered, using metrics such as throughput, latency, packet loss and jitter. The framework enables reconfigurable, on-demand tradeoff analysis to derive optimal solutions. Utilizing this framework, the authors present findings for the network performance impact of running a Virtual Battlespace 3 (VBS3) training exercise on the 29 Palms network.