Parallel implementation for dynamic mesh optimization on distributed computer system.

The large scale computer system provides a high performance platform for engineering applications. Mesh generation is the basis for numerical simulation for computing science, which heavily relies on user’ experience and may not match the demand of application itself. Dynamic mesh optimization according to the features of application is an automatic and smart way during the parallel simulation. The shared-memory parallelism on a standalone computer is the main strategy for mesh generation which is regarded as a pre-processing of simulation. However, it is not suitable for dynamic mesh optimization which is involved in parallel simulation commonly on distributed computer system. In this paper, the parallel implementation on distributed system is established. A two-stages method was proposed for mesh parallel editing to make new mesh objects parallel consistent. The parallel repartition-migration method was devised to recover the load balance for numerical simulation after mesh editing. Parallel performance test shows that parallel mesh editing is a communication dense procedure and so the parallel efficiency drops as the computing cores increase.