Ice-Sheet Meshes and Biconnectivity Ian Bogle (RPI), Karen Devine - - PowerPoint PPT Presentation

ice sheet meshes and biconnectivity
SMART_READER_LITE
LIVE PREVIEW

Ice-Sheet Meshes and Biconnectivity Ian Bogle (RPI), Karen Devine - - PowerPoint PPT Presentation

Ice-Sheet Meshes and Biconnectivity Ian Bogle (RPI), Karen Devine (SNL), Mauro Perego (SNL), Siva Rajamanickam (SNL), George Slota (RPI) Motivation


slide-1
SLIDE 1

Ice-Sheet Meshes and Biconnectivity

Ian Bogle (RPI), Karen Devine (SNL), Mauro Perego (SNL), Siva Rajamanickam (SNL), George Slota (RPI)

slide-2
SLIDE 2

Motivation

https://insideclimatenews.org/sites/default/files/styles/icn_full_wrap_wide/public/getz-ice-shelf_jeremy-harbeck-nasa.jpg?itok=D1uLzpFu

slide-3
SLIDE 3

Background

  • Biconnectivity
  • Articulation points
  • Meshes
  • Potential articulation points
  • “Double hinges”
slide-4
SLIDE 4

The Ice-Sheet problem

  • Modeling Antarctic Ice-Sheet
  • “Grounding”
  • Degenerate Features
  • Similar to Biconnectivity
slide-5
SLIDE 5

Label Propagation

  • Three pieces:
  • Grounding
  • Articulation point heuristic
  • No false negatives
  • Propagation rules
  • Propagate grounding throughout the mesh
slide-6
SLIDE 6

Labels

  • Each node has a unique vertex ID
  • Labels contain four vertex IDs
  • Two for “grounded” nodes
  • Two for senders
  • Labels can be
  • Empty, no vertex IDs
  • Half-full, one pair of vertex IDs
  • Full, two pairs of vertex IDs
  • Nodes with full labels are considered grounded
slide-7
SLIDE 7

Basic Propagation Rules

  • Potential Articulation Points
  • Pass own ID if grounded
  • Only pass label to neighbor once
  • Can pass label too early
  • All other nodes
  • Pass whole label unconditionally
  • Only pass labels from nodes whose labels have changed recently.
  • Since nodes only change twice, we visit each node at most twice.
slide-8
SLIDE 8

Example

slide-9
SLIDE 9

Incomplete Propagation

  • Due to potential articulation point restriction
  • Can be detected and fixed
  • Very rare in ice-sheet cases
slide-10
SLIDE 10

Results vs. Biconnectivity Algorithms

Ice-sheet resolution BCC-BFS BCC-Color Label Propagation 16 km 0.0163 (s) 0.0841 (s) 0.0046 (s) 8 km 0.0483 (s) 0.7728 (s) 0.0196 (s) 4 km 0.1912 (s) 7.6713 (s) 0.0834 (s) 2 km 0.7199 (s) 54.821 (s) 0.3395 (s) 1 km 3.3271 (s)

  • 1.3904 (s)
slide-11
SLIDE 11

Generalizing to Biconnectivity

  • Need grounding and articulation point heuristic
  • Any two neighbors for grounding
  • LCA Heuristic
  • Propagation rules work fine
  • Only finds one biconnected component per propagation
  • Iterative
slide-12
SLIDE 12

Lowest Common Ancestor Heuristic

slide-13
SLIDE 13

Distributed Memory

  • The original problem exists in distributed memory
  • The propagation rules still work
  • Ice-sheet case is simple to distribute
slide-14
SLIDE 14

Distributed Memory

  • Biconnectivity version is more difficult
  • Distribute LCA Heuristic
  • Strategically pick neighbors for the 2-Clique
  • Easier using Trilinos
slide-15
SLIDE 15

Current State

  • Working on an MPI implementation
  • Work-efficient (?) distributed biconnectivity algorithm
  • Working to publish in ISC