Runtime Optimisation Approaches for a Real-Time Evacuation Assistant

* Final gross prices may vary according to local VAT.

Get Access

Abstract

This paper presents runtime optimisation approaches for a real-time evacuation assistant. The pedestrian model used for the forecast is a modification of the centrifugal force model which operates in continuous space. It is combined with an event driven route choice algorithm which encompasses the local shortest path, the global shortest path and a combination with the quickest path. A naive implementation of this model has the complexity of O(N 2), N being the number of pedestrians. In the first step of the optimisation the complexity is reduced to O(N) using special neighbourhood lists like Verlet-List or Linked-Cell commonly used in molecular dynamics. The next step in this optimisation process is parallelisation on a multicore system. The Message Passing Interface (MPI) and Open Multi-Processing (OpenMP) application programming interfaces are used to this extend. The simulation is performed on the Juropa cluster installed at the Jülich Supercomputing Centre. The speedup factors obtained are ~10 for the linked-cells, ~4 for 8 threads and ~3 for the parallelisation on 5 nodes using a static domain decomposition.