Five papers address interrelated topics in parallel computing and the use of adaptive, unstructured computational grids. The titles of the papers and their general subject matters are the following:
- "Parallelization of a Dynamic Unstructured Application using Three Leading Paradigms" compares the parallelization of a dynamic unstructured-mesh-adaptation algorithm by use of a message-passing, a shared-memory, and a multithreading implementation on three supercomputers. It is concluded that multithreaded systems offer tremendous potential for quickly and efficiently solving some of the most challenging real-life problems on parallel computers.
- "Self-Avoiding Walks Over Two-Dimensional Adaptive Unstructured Grids" presents an approach, based on a special class of self-avoiding walks, for partitioning computational loads among parallel processors in cases of adaptive unstructured grids.
- "A Dynamic Load Balancing Framework for Unstructured Adaptive Computations on Distributed-Memory Multiprocessors" describes the JOVE computer program, which balances loads among parallel processors according to a global view each time an unstructured computational grid is adapted.
- "Efficient Load Balancing and Data Remapping for Adaptive Grid Calculations" presents an implementation and integration of all major components of a load-balancing strategy, including the interface between a parallel mesh-adaption code and a data-remapping module.
- Dynamic Load Balancing for Adaptive Meshes using Symmetric Broadcast Networks" demonstrates that the performance of a novel load-balancing algorithm that utilizes symmetric broadcast networks is comparable to that of a previously developed global load-balancing algorithm.
This work was done by Rupak Biswas and Leonard Oliker of Ames Research Center, Gerd Heber and Guang R. Gao of the University of Delaware, Andrew Sohn of New Jersey Institute of Technology, Horst D. Simon of Lawrence Berkeley National Laboratory, and Sajal K. Das and Daniel J. Harvey of the University of North Texas. To obtain copies of the papers, access the Technical Support Package (TSP) free on-line at www.nasatech.com/tsp under the Information Sciences category.