CSC Homepage

Masters Project Defense by Ashley Nicole Zebrowski


Presentation Title: " The BL-Octree: An Efficient Data Structure for Discretized Block-Based Adaptive Mesh Refinement"

Committee:
  • Dr. Gabrielle Allen (Major Professor)
  • Dr. Steven Brandt
  • Dr. Shantenu Jha
  • Dr. Supratik Mukhopadhyay
Date: July 06, 2011
Time: 9:00 AM
Location:279, Coates Hall

Abstract:
Adaptive mesh refinement (AMR) algorithms require efficient structures to store and operate upon volumetric grid data that will be used to distribute work among processes. When inefficient algorithms are used, AMR-enabled software may have its scaling negatively impacted as the set operations crucial for AMR calculations become an artificial bottleneck. This paper presents a variable-resolution octree-based structure which efficiently handles the set operations needed for AMR operations such as calculating process communication schedules and buffer zones. Preliminary results indicate a marked improvement in execution times when large amounts of volumetric data are processed, but at the price of a structure creation and traversal penalty for smaller amounts of data. The BL-octree ("blocktree") data structure presented in this paper will therefore be of use in massively parallel AMR environments, where the cost of distributing AMR workloads across many tens of thousands of processes or more can be reduced.

All are invited.

LSU Home ""

Department of Computer Science
Louisiana State University
298 Coates Hall
Baton Rouge, LA 70803
Phone: (225) 578-1495
Fax : (225) 578-1465