Grid partitioning versus domain decomposition: A comparison of some industrial problems on workstation clusters (Q1925098)

In the field of nonlinear differential equations two main techniques of solving problems in parallel can be observed: data partitioning techniques on the one hand and domain decomposition methods on the other hand. Data partitioning techniques are somewhat direct parallelization strategies whereas domain decomposition methods are more indirect in that they are characterized by an additional iteration to enforce compatibility at inner boundaries. This paper gives a survey of both techniques and compares their performances in industrially relevant problems.