Performance considerations for In-Situ Adaptive Manifolds for turbulent combustion modeling

Manifold-based combustion models reduce the computational cost of turbulent

combustion simulations by projecting the thermochemical state onto a lower-

dimensional space and deriving manifold equations for the thermochemical state that

can be solved separately from the flow solver. Traditionally, the solutions of the manifold

equations have been precomputed and tabulated, and the precomputation cost and

tabulation memory requirements have limited traditional manifold-based models to

simple single ‘mode’ combustion processes. With the recently developed approach

termed In-Situ Adaptive Manifolds (ISAM), solutions of the manifold equations are

instead computed ‘on-the-fly’ and stored for efficient reuse using In-Situ Adaptive

Tabulation (ISAT), allowing for more general manifold-based models for multi-modal

combustion. However, ISAM suffers from two computational bottlenecks. First, the cost

of the manifold solver is the largest computational cost and its minimization would

accelerate the algorithm. Second, to avoid redundant manifold solves across parallel

processes, the ISAT database should be either distributed or synchronized across

processes. These two performance considerations have been profiled, and various

performance improvement strategies are discussed and evaluated.