Large Scale Parallel Simulation Optimization on a Network of Heterogeneous Workstations

Abstract

This paper analyses results from large scale parallel computations obtained from a distributed approach which uses numerical simulation and optimization techniques to automatically find remediation solutions to a hypothetical contaminated aquifer. The remediation strategy is based on withdrawal, which requires the removal of contaminated groundwater from the aquifer by pumping. The design of the remediation system involves the choice of the number of wells to be installed, their locations and pumping rates, with the goal of maximizing the amount of contaminant extracted, while minimizing the cost of the system. The optimization strategy adopted, a Genetic Algorithm (GA), requires a large number of calls to the numerical simulation model of the aquifer, which tends to be computationally expensive. To overcome this drawback, the numerical simulations are executed in parallel, using a network of heterogeneous workstations. Performance of the parallel approach, using 95 and 190 cores, is analysed.