1 / 9

Toward Automatic Parallel Adaptive Mesh Refinement

Toward Automatic Parallel Adaptive Mesh Refinement. Scott H. Hawley*, Matthew W. Choptuik* U *University of Texas at Austin * U University of British Columbia shawley@einstein.ph.utexas.edu. Credits: Manish Parashar , James C. Browne, Paul Walker,

Download Presentation

Toward Automatic Parallel Adaptive Mesh Refinement

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Toward Automatic Parallel Adaptive Mesh Refinement Scott H. Hawley*, Matthew W. Choptuik*U *University of Texas at Austin *UUniversity of British Columbia shawley@einstein.ph.utexas.edu Credits: Manish Parashar, James C. Browne, Paul Walker, Shyamal Mitra, Robert Marsa, Mijan Huq, Dae-Il Choi

  2. Motivation When we model physical phenomena using finite-difference approximations of partial differential equations… • For fixed local accuracy, required resolution may vary widely in space and time • Resolution requirements may not be known a priori • Adaptive Mesh Refinement (AMR) Even with the utility AMR provides, a code must be parallelizable to take advantage of modern computing machinery

  3. Motivation, cont’d • AMR and parallel processing are desirable, but both present challenges which may be prohibitive for many researchers • Investigate environments in which AMR and parallelism are provided automatically

  4. Paradigm • Almost all details of AMR and parallelism hidden from user • Provide unigrid routines • Specify • maximum # of levels • truncation error tolerance for regridding • clustering efficiency • Entire AMR driver generated automatically • User selects “AMR: On” (someday soon)

  5. Build Around GrACE GrACE provides structures for AMR and parallelization Goal: Make GrACE features easily accessible to end user Provide: • Generic Driver (“Your code here”) • Output support • Supplemental Documentation (“How to…”) • Link to RNPL

  6. Rapid Numerical Prototyping Language (RNPL) Marsa & Choptuik • Minimal development time • Specify: • Initial Data • Boundary Conditions • Finite Difference Equations • Examples: • Pedagogy: Scalar wave in IEF coordinates • Boson star simulations • Generate framework for fluid codes • Easily used to write Cactus Thorns

  7. Coincident Goals Both this effort and Cactus seek automatic, parallel AMR in the very near future • Cactus needs to deliver AMR • Generic GrACE driver could be run as a Cactus Thorn • Maybe “The” Cactus AMR Thorn • Problems my group want to solve • My dissertation: Accretion disk theory within IMSO (Scott needs to land a job…) • Others: Gamma-ray burst models, Multi-D critical phenomena, and much more! David Neilsen, Jason Ventrella, Ethan Honda, Scott Noble • Cactus better connected with a wide variety of computing support than we can provide on our own

  8. Needs • Visualization of AMR data • Interactive tool for daily use • Not necessarily “flashy” or “high-performance” • Inexpensive • Curvilinear coordinate systems • Animation • Grid-grid operations (+,-,*,/,etc) • Easy for user to add new functionality (filters, parameters) • Efficient Collaboration • Daily e-mail is not interactive enough to achieve short turn-around

More Related