< Back to previous page


Reducing computational cost of large-scale simulations using opportunistic model approximation

Book Contribution - Book Abstract Conference Contribution

We present a dynamic model approximation strategy that allows to significantly increase computational efficiency of the simulation while maintaining proper validity. This can be used to effectively overcome the scalability constraints in state-of-the-art simulation frameworks for testing and validating large-scale systems. The method that we present leverages information theory metrics to measure the possible contribution of sub-areas in the simulation to the global behavior. This allows us to opportunistically approximate low-contributing areas and as a result decrease the computational cost of the simulation. We present a basic traffic-simulation use-case, implemented in the Acsim simulator to validate the proposed method and are able to achieve a 33% reduction of the computational cost. Furthermore, we analyze our proposed method from a more theoretical perspective.
Book: 2019 Spring Simulation Conference (SpringSim), 29 April-2 May, 2019, Tucson, Arizona, USA
Number of pages: 12
Publication year:2019
Keywords:P1 Proceeding
Authors from:Higher Education