The chemical task in internal combustion engine simulations concerns with the solution of a non-linear stiff system of Ordinary Differential Equations (ODEs) per each cell of a discretization grid representing engine geometry. The computational cost of the above task, when a detailed kinetic scheme is used, is dominating in engine simulations. Due to local physical-chemical conditions, each system of ODEs is characterized by local numerical properties (such as stiffness), therefore local adaptive solvers are usually employed for its efficient solution. We developed an MPI-based combustion parallel solver for efficient solution of the chemical task in engine simulations within parallel environment. In this context, we propose a cell distribution based on a dynamic load balancing algorithm, using a strategy which preserves contiguousness of the computational grid cells. Efficiency of our approach is shown for parallel simulations of realistic Diesel engines, when different sizes of the discretization grid and different operative conditions of the engine are used.
Dynamic Load Balancing for High-Performance Simulations of Combustion in Engine Applications
Laura Antonelli;Pasqua D'Ambra
2011
Abstract
The chemical task in internal combustion engine simulations concerns with the solution of a non-linear stiff system of Ordinary Differential Equations (ODEs) per each cell of a discretization grid representing engine geometry. The computational cost of the above task, when a detailed kinetic scheme is used, is dominating in engine simulations. Due to local physical-chemical conditions, each system of ODEs is characterized by local numerical properties (such as stiffness), therefore local adaptive solvers are usually employed for its efficient solution. We developed an MPI-based combustion parallel solver for efficient solution of the chemical task in engine simulations within parallel environment. In this context, we propose a cell distribution based on a dynamic load balancing algorithm, using a strategy which preserves contiguousness of the computational grid cells. Efficiency of our approach is shown for parallel simulations of realistic Diesel engines, when different sizes of the discretization grid and different operative conditions of the engine are used.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.