Proceedings of the 2016 Winter Simulation Conference T. M. K. Roeder, P. I. Frazier, R. Szechtman, E. Zhou, T. Huschka, and S. E. Chick, eds. ENERGY CONSUMPTION OF DATA DRIVEN TRAFFIC SIMULATIONS SaBra Neal Michael Hunter Richard Fujimoto Computational Science and Engineering Civil and Environmental Engineering Georgia Institute of Technology Georgia Institute of Technology Atlanta, GA 30332, USA Atlanta, GA 30332, USA ABSTRACT Dynamic Data-Driven Application Systems (DDDAS) implemented on mobile devices must conserve energy to maximize battery life. For example, applications for online traffic prediction require use of real- time data streams that drive distributed simulations. These systems involve embedding computations in mobile computing platforms that establish the state of the system being monitored and collectively predict future system states. Understanding where energy consumption takes place in such systems is vital to optimize its use. Results of an empirical investigation are described that measure energy consumption of aspects such as data streaming, data aggregation, and traffic simulation computations using different modeling approaches to assess their contribution to overall energy consumption. 1 INTRODUCTIONS Energy consumption is an on-going concern in mobile and embedded computing systems powered by the device’s battery. With the growing use of real-time data for traffic prediction applications one must understand tradeoffs between energy consumption for communications and computations under certain performance and accuracy constraints in order to ensure effective operation. For example, question might concern the approach used to model the system and the amount and frequency with which data should be collected to drive the simulation computations. This information is necessary to develop power and energy aware techniques to optimize energy use. Dynamic Data Driven Application Systems (DDDAS) allow simulations to incorporate real-time or online data in order to drive the simulation system to produce predictions that can be used to aid measurements or optimize system operation (Darema 2004). DDDAS applications may be embedded within the physical system being monitored or optimized in order to utilize real-time data near the source of the data. For example, embedded traffic simulations may be part of a sensor network where real-time traffic data is used as input to drive transportation simulations. Monitoring ecological development, forest fires, and tracking multiple targets in an ad hoc sensor network are examples where a DDDAS system might be embedded within the physical system (Rodríguez et. al. 2009; Schizas and Maroulas 2015). In situations where battery-powered mobile devices are used as the DDDAS platform energy consumption by DDDAS computations and communications is an important issue. Ad-hoc distributed simulation systems have been proposed for applications such as data-driven distributed traffic network simulations (Fujimoto et. al. 2007). These distributed simulations may be implemented in sensor networks where sensor data is communicated to distributed simulation processes within close physical proximity. In a transportation application each simulation is responsible for using the sensor data to make future state predictions about a portion of the traffic network and exchange current and predicted state information with other simulations. These simulations collectively predict the future state of the traffic network as a whole. Individual simulations may be in close proximity of the 978-1-5090-4486-3/16/$31.00 ©2016 IEEE 1119