Scalable Recursive Clustering
The scalability of the Fire Dynamics Simulator (FDS) with respect to massively parallel computers largely relies on the efficient solution of the pressure equation which is closely coupled with the computation of all other thermodynamic quantities. The current FDS pressure solver is based on the mesh-wise use of direct Fast Fourier Transformations (FFT) which are iteratively coupled with an additional averaging process along internal mesh boundaries. This purely local way of proceeding can cause difficulties concerning the reliable approximation of global data dependencies encompassing the whole computational domain which in turn may impair the effective use of multi-mesh computations whilst retaining a high level of accuracy at the same time.
As an alternative approach the solver package Scalable Recursive Clustering (ScaRC) is under development consisting of a selection of different iterative solution techniques of domain decomposition and multi-grid type which belong to the most efficient and robust solvers for huge systems of equations. Based on elaborated numerical combinations of local and global mechanisms for capturing the overall physical effects, ScaRC has already proven considerably enhanced scalability and accuracy properties for a variety of different test and verification cases. This article is intended to give an overview of the underlying concepts as well as the current state of development accompanied by some illustrative numerical test examples and comparisons with the current FFTsolver.