Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 95
PROCEEDINGS OF THE SECOND INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GRID AND CLOUD COMPUTING FOR ENGINEERING
Edited by:
Paper 95

Parallel Real Time Computation of Large Scale Pedestrian Evacuations

B. Steffen, U. Kemloh, M. Chraibi and A. Seyfried

Forschungszentrum Jülich GmbH, Jülich Supercomputing Centre, Germany

Full Bibliographic Reference for this paper
B. Steffen, U. Kemloh, M. Chraibi, A. Seyfried, "Parallel Real Time Computation of Large Scale Pedestrian Evacuations", in , (Editors), "Proceedings of the Second International Conference on Parallel, Distributed, Grid and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 95, 2011. doi:10.4203/ccp.95.95
Keywords: evacuation, simulation, real time computation, force based model, parallelization, visualization.

Summary
This system consists of an large number of sensors, a 25 node cluster computer for running the simulation, and a front-end system with good graphics. The central problem is the computation of a sufficiently accurate prediction much faster than real time. For this, the generalized centrifugal force model of pedestrian motion is used. This model works with repulsive forces between nearby persons that depend on the difference in speed over distance. This model is enhanced with a linked list ordering of persons to enhance the speed by utilizing the strong locality of the force model. It is parallelised over one multicore processor or one two-processor node of the cluster using OpenMP. It will be parallelised over many nodes using MPI with a domain decomposition scheme. The code can now handle 1200 persons on one processor with sufficient speed, but requires further optimization. The calculation of mutual distances of persons is the most time consuming part. Better use of the force localization, enhanced distance calculation and a faster ODE solver will improve the sequential speed, and better load balancing the parallel efficiency. Jointly, these improvements are likely to provide the required speed of calculation.

A further important aspect is the rapid visualization of results. The staff has only very few minutes to understand the problems of 15 minutes of evacuation time. For this, a number of tools are to be prepared for testing during installation. Experience will demonstrate the kind of visualization that is most intuitive and helpful. If successful, it will be a basis for further systems using simulations to aid crowd management

References