Proceedings of the 2011 Winter Simulation Conference S. Jain, R.R. Creasey, J. Himmelspach, K.P. White, and M. Fu, eds. STATISTICAL ISSUES IN AD HOC DISTRIBUTED SIMULATIONS Ya-Lin Huang Richard Fujimoto Wonho Suh Michael Hunter Georgia Institute of Technology Computational Science and Engineering Georgia Institute of Technology Civil and Environmental Engineering Atlanta, GA 30332, USA Atlanta, GA 30332, USA Christos Alexopoulos Georgia Institute of Technology Industrial and Systems Engineering Atlanta, GA 30332, USA ABSTRACT An ad hoc distributed simulation is a collection of online simulators embedded in a sensor network that communicate and synchronize among themselves. Each simulator is driven by sensor data and state pre- dictions from other simulators. Previous work has examined this approach in transportation systems and queueing networks. Ad hoc distributed simulations have the potential to offer greater resilience to fail- ures, but also raise a variety of statistical issues including: (a) rapid and effective estimation of the input processes at modeling boundaries; (b) estimation of system-wide performance measures from individual simulator outputs; and (c) correction mechanisms responding to unexpected events or inaccuracies of the model itself. This paper formalizes these problems and discusses relevant statistical methodologies that allow ad hoc distributed simulations to realize their full potential. To illustrate one aspect of these meth- odologies, an example concerning rollback threshold parameter selection is presented in the context of managing surface transportation systems. 1 INTRODUCTION Uses of online simulations are emerging as the needs for solving problems in operational systems in real- time become more and more common. Many real-world problems are sufficiently complex that analytical solutions may not exist or may be too complicated to be solved efficiently. Simulations offer an alternate approach where a model that mimics the underlying physical system is created and driven by sensor data and problem-specific configurations. A successful online simulation solution must satisfy real-time- related constraints and be equipped with the ability to capture the dynamics of the physical system. Online simulations are also referred to as dynamic data-driven application systems (DDDAS), symbi- otic simulations (Fujimoto et al. 2002), and cyber-physical systems in the literature. They have been widely studied and applied to various science and engineering disciplines for a diverse collection of pur- poses (Davis 1998). A typical application is to optimize the operation of a physical system. For example, in an emergency situation alternate evacuation scenarios may be modeled and evaluated in order to mini- mize evacuation time. The evacuation plan may need to adapt as the evacuation evolves with new unfore- seen events arising. Other online simulation applications include path planning for unmanned aerial vehi- cles (Kamrani and Ayani 2007), parameter tuning for computer networks (Ye et al. 2008), management of 612 978-1-4577-2109-0/11/$26.00 ©2011 IEEE