Parallelization of Isogeometric Analysis
on Memory Distributed Computing Platforms

Daniel Rypl, Bořek Patzák

Department of Mechanics
Faculty of Civil Engineering
Czech Technical University in Prague
Thákurova 7, 166 29 Prague, Czech Republic



Abstract:

Isogeometric analysis has been recently introduced as a viable alternative to the standard, polynomial-based finite element analysis. Similarly to the finite element method, isogeometric solution of complex engineering problems may lead to computationally very demanding analysis, demands of which can be alleviated by performing it in a parallel computing environment. This paper presents how the isogeometric analysis based on NURBS can be parallelized on memory distributed computing platforms. While the actual parallelization of the isogeometric computational code resembles methodologically very much the parallelization of the finite element code, the construction of the appropriate domain decomposition of the isogeometric mesh is rather complicated compared to the partitioning of the finite element mesh. The paper introduces a new methodology, how to construct a weighted dual graph of a two-dimensional NURBS-based isogeometric mesh that can be decomposed by standard graph-based partitioning approaches. The application of the proposed strategy is illustrated on a simple two-dimensional example.