Mesh Optimization Algorithms for Parallel Computing with MESQUITE

Sandia National Laboratories Computer Science, 2004-05

Liaison(s): Patrick Knupp
Advisor(s): Melissa O’Neill
Students(s): Dominik Slusarczyk (PM), Elisa Celis, John Hicks, Yu-Min Kim

This Clinic project extended the MESQUITE mesh smoothing toolkit developed by Sandia National Laboratories to operate on a distributed processing cluster. Parallel smoothing requires efficient partitioning of meshes into subparts, correct smoothing of those subparts, and effective cross-cluster synchronization during and after computation. The project drew on existing research in the field of distributed mesh smoothing and on established tools, including MESQUITE itself, the Zoltan partitioning toolkit, and the MPI toolkit for distributed computing. Distributed computation would be pointless without speedup over ordinary single-CPU computation, so the team has also developed and deployed performance analysis methods which have inspired further optimizations to the code.