High Performance Programming Models for Monte Carlo Methods

Abstract

Complexity of modern supercomputers is rising steadily, especially with the introduction of heterogenous and distributed computing resources. This complexity also affects the available parallel programming models. This work focuses on this issue in the context of the Monte Carlo method.

In this work we propose two applications of the Monte Carlo method. The first one comes from the field of hydrology where it is used to model uncertainty in the Rainfall-Runoff models. This result can improve the decision making process of the local governments in case of an incoming flood. This application is run on a high performance computing infrastructure which allows fast delivery of precise results. The method is integrated in a custom simulation framework optimized for parallel execution. Its outputs are integrated in an web-based interface provided by the Floreon+ system for easy access. The outputs enhance the prediction of a water discharge and inundation areas by introducing confidence intervals to their output.

The second application comes from the field of traffic navigation and optimization. The Monte Carlo method is used to estimate travel time distribution on a given path using the Probabilistic Time-Dependent Routing algorithm. This application is integrated in an experimental on-line server side traffic navigation service which can be used to optimize traffic in the context of future Smart cities. The optimized version of this algorithm allows the service to provide optimal routes to a large number of cars driving through a given region.

For both applications we document their design, implementation, optimization and deployment on a high performance computing infrastructure. Computational and performance experiments were executed with both applications, their results are presented in the corresponding chapters.