Performance of stabilized explicit time integration methods for parallel air quality models

  • Authors:

  • Anurag Srivastava;John C. Linford;Adrian Sandu

  • Affiliations:

  • Virginia Polytechnic Institute and State University, Blacksburg, VA;Virginia Polytechnic Institute and State University, Blacksburg, VA;Virginia Polytechnic Institute and State University, Blacksburg, VA

  • Venue:
  • SpringSim '07 Proceedings of the 2007 spring simulation multiconference - Volume 2
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Comprehensive air quality models (AQMs) describe the fate and transport of atmospheric chemical constituents associated with the gas and aerosol phases. Air quality studies involve multi-scale multi-physics simulations which follow the evolution of ~ 107 -- 108 variables on relatively long time scales. Air quality simulations are therefore computationally intense and require efficient parallelization to make them practical. The goals of this work are twofold. First we study the use of Runge Kutta Chebyshev (RKC) time integration of the advection-diffusion equations in AQMs. RKC are stabilized explicit time integration methods which are appropriate for the time integration of mildly stiff diffusion terms. Second, we assess the benefits of a tiled domain decomposition scheme for parallelization of AQMs. This approach is feasible due to the reduced data dependencies associated with explicit time stepping. Numerical results indicate that RKC methods have the necessary stability properties to handle turbulent diffusion. Moreover, the parallelization based on tiled domain decomposition with explicit time stepping shows a much-improved parallel performance when compared to the parallelization of implicit time stepping schemes. Our benchmarks also highlight a possibly significant bottleneck in the I/O system of an air quality model. This research demonstrates the viability of RCK time stepping methods and motiviates future research into parallel I/O.