Dynamic Load Balancing and Data Migration for Adaptive Numerical Methods

Project Description

In this project, we will design and implement algorithms for dynamic load-balancing on homogeneous and heterogeneous parallel computers. In particular, we will develop load-balancing algorithms that support adaptive mesh refinement. Key steps in this research include:
  • Develop cost models and measurements for processor loads and data migration in adaptive methods.
  • Develop and implement dynamic load-balancing algorithms that take advantage of octree data structures while limiting migration costs.
  • Investigate trade-offs involved with load-balancing hierarchical meshes and distributing octree data structures across parallel computers.
  • Implement efficient octree migration strategies.
  • Personnel

  • Karen D. Devine, Sandia National Laboratories, Albuquerque, NM
  • Gary L. Hennigan, New Mexico State University, Las Cruces, NM
  • Related Projects

  • Chaco: Static graph partitioning algorithms for general parallel applications. (Funded by DOE MICS Division office.)
  • MPSalsa : Massively parallel finite element code for the simulation of complex chemically reacting flows. (Funded in part by DOE/Office of Scientific Computing.)
  • Last Updated: October 15, 1997
    WWW Administration (www-admin@www.cs.sandia.gov)
    Karen Devine (kddevin@cs.sandia.gov)

    Page 1