Dynamic Multicore Processing for Pandemic Influenza Simulation

Abstract

Pandemic simulation is a useful tool for analyzing outbreaks and exploring the impact of variations in disease, population, and intervention models. Unfortunately, this type of simulation can be quite time-consuming especially for large models and significant outbreaks, which makes it difficult to run the simulations interactively and to use simulation for decision support during ongoing outbreaks. Improved run-time performance enables new applications of pandemic simulations, and can potentially allow decision makers to explore different scenarios and intervention effects. Parallelization of infection-probability calculations and multicore architectures can take advantage of modern processors to achieve significant run-time performance improvements. However, because of the varying computational load during each simulation run, which originates from the changing number of infectious persons during the outbreak, it is not useful to us the same multicore setup during the simulation run. The best performance can be achieved by dynamically changing the use of the available processor cores to balance the overhead of multithreading with the performance gains of parallelization.

Figure 1.

Simulation run time versus number of cores for different thresholds for activating multithreading

Figure 2.

Simulation run time versus number of cores for an intervention scenario where schools close at day 10 and later reopen at day 20.

Figure 3.

Simulation run time versus number of cores for an intervention scenario where schools close at day 10 and later reopen at day 20.

Figure 4.

Simulation run time versus number of cores for two simultaneous simulation jobs running on the same machine.