Flow and combustion models are being used to evaluate new designs and retrofit options for various industrial combustion systems. Combustion models being used today are often very modular and, since they apply serial algorithms, require long run times to produce results. It is common for solutions to take several days, and the use of finite rate chemistry and Lagrangian based particle models can lengthen run times to a week or more. The modularity of these methods makes them candidates for parallel computing.
This paper presents results for a distributed computing algorithm using the PVM software, which is applied to the finite rate chemistry and particle transport modules. It is based on a master-slave algorithm in which the master doles work to a number of independent processors. A load balancing scheme is used to account for the variability in the time the slaves complete their work.
PVM was successfully used for parallel computations in the finite rate chemistry and particle modules. Significant speedups were found for both modules, but the work clearly indicates the need to control granularity and the need to optimize the algorithm specifically for the processors being used. Future work is planned to improve the algorithms presented here as well as extending the work to other parts of the combustion model.