Efficient parallel generation of partitioned, unstructured meshes

In this paper we introduce a method for generating unstructured meshes in parallel which are partitioned in a ‘good’ way. When solving a partial differential equation on a parallel distributed memory machine, the mesh should be decomposed so that the communication requirement of the numerical solver is minimised and also the amount of work to be performed on each processor is approximately equal.

Most previous work in this area has concentrated on partitioning a mesh that has already been generated. We introduce a method which actually generates the partitioned mesh in parallel whilst producing a good quality decomposition and compare this method with other approaches.