Providing adequate software support for unstructured grid generation codes is challenging even for sequential implementations. In parallel implementations, program complexity increases by an order of magnitude due to dynamic, data-dependent, irregular computation and communication requirements. Even more challenging is the implementation of parallel methods that maintain the same quality guarantees as their sequential counterparts. In this paper we present a data migration run-time system, the Mobile Object Layer, for implementing load balancing heuristics for parallel adaptive unstructured mesh computations. We also argue that the Mobile Object Layer (MOL) simplifies this formidable task. We present preliminary performance data from a parallel mesh generator, based on the Constrained Delaunay Triangulation (CDT), which suggest that the flexibility and general nature of our data-migration approach does not cause undue overhead.
展开▼