DEX: Dexterous Data Explorer

1
DEX Dexterous Data Explorer

• Kurt Stockinger, John Wu
• (Scientific Data Management-Group)
• Wes Bethel, John Shalf
• (Visualization-Group)
• Berkeley Lab
• Scientific Data Management Center All Hands
  Meeting
• March 2005
• Salt Lake City, Utah

2
Motivation Scientific Data Exploration

• Combustion
• Tracking species of molecules through chemical
  reaction networks
• Studying flame fronts helps better understand
  efficient combustion
• Approach
• Select data that characterizes flame front based
  on user condition
• Perform visual analysis
• Astrophysics
• Tracking of mass fractions of different chemical
  species

3
Traditional Approaches

• Database access methods
• E.g. Bitmap indices for multi-dimensional data
  analysis
• Visualization frameworks
• Isosurface extraction based on pre-defined values
• Applications require a combination of both
  worlds

4
DEX Dexterous Data Explorer

• Interactive visualization of large scientific
  data sets
• Uses a novel combination of efficient query
  technology and visualization infrastructure
• FastBit Query Engine Bitmap indices for
  accelerating queries on large data sets for
  identifying regions of interest
• Visualization pipeline for generating 3D images
  of abstract data results
• DEX provides new functionality that is not
  present in any visualization analysis software
• Perform interactive, multi-dimensional queries to
  refine regions of interest that are later used as
  input to analysis or visualization
• By displaying the resulting regions of interest,
  application scientists can quickly identify
  characteristic features of their data

5
Simplified Visualization Pipeline
FastBit
Query
Data
Visualization Toolkit(VTK)
3D visualization of a Supernova explosion
6
Performance Analysis

• Experimental Setup
• 2.8 GHz Intel Pentium IV with 2 GB RAM
• SCSI RAID disk
• Performance comparison of DEX with isosurface
  algorithms in VTK
• Two data sets
• Combustion
• 2.7 million data points with 10 features
• Astrophysics
• 13.8 million data points with 6 variables
• Bitmap Index
• WAH-compressed, range-encoded bitmap index
• 100 bins

7
Performance Results with Scientific Data
Visualization of results is similar to isosurface
extraction
DEX is on average a factor of three to four
faster than best isosurface algorithm of VTK
8
Future Work

• Reduce bottleneck in vis-pipeline
• Region growing is small fraction of VTK rendering
• Multi-resolution analysis
• AMR data
• HDF5 (we currently support netCDF HDF4)

9
More Information

• Bitmap Indices
• Google Kurt Stockinger
• Google John Wu
• DEX
• Google DEX bitmap
• http//www-vis.lbl.gov/Research/Dex/