Accelerating Veins Simulations by Utilizing Task Parallelism on a HPC Cluster without Introducing Major Inaccuracies

Abstract

The evaluation of large-scale vehicular networks has proven difficult. The researcher has to either do a lot of experiments, reduce simulation accuracy or endure simulation runtimes prohibiting a reasonable investigation. We present a framework that allows preserving simulation accuracy while maintaining reasonable simulation times by cutting the to-be-investigated area into individual tiles. Instead of running on a single CPU core we are able to run the simulation on hundreds of HPC servers in parallel. This yields a speedup of several orders of magnitude, allowing the simulation of entire cities. We validate our approach by investigating different access-point distributions for the city of Ingolstadt.