Interactive Visualization of Very Large Multiresolution Scientific Data Sets

1
Interactive Visualization of Very Large
Multiresolution Scientific Data Sets
Dan Bergeron, Computer Science Andrew Foulks,
Computer Science Jimmy Raeder, Physics
2
Presentation Overview

• Project aspirations
• MR/AR Time/Space multiresolution
• Out-of-core interactive visualization
• Case study
• MHD
• Simulation of solar winds
• Project reality
• collaboration means compromise
• delivery is at least as important as function

3
Interactive Visualization Model

• Generates multiresolution data (in both spatial
  and temporal domains)
• Initial view is at a coarse enough level to
  support interactivity (depends on platform)
• Zoom into spatially and/or temporally focused
  view at higher resolution
• where the data is "interesting", and
• where the data has high error
• Goal memory demand stays constant

4
Case Study Overview

• Challenges of visualizing simulation data
• Focus on unsteady MHD simulation
• Application framework
• Time Series Data
• Multi/Adaptive resolution techniques
• Error model
• STAR data
• Space Time Adaptive Resolution data

5
Solar Wind Simulation

• Models interaction between solar wind and Earths
  magnetosphere
• Simulation records magnetic field, particle
  velocity, density and more
• Data is a 3D time series
• Data points sampled on a structured grid
• 87 time steps, total data size is 15GB

6
Solar Wind Unsteady Flow
392 x 112 x 112
196 x 56 x 56
Note color is mapped to particle speed
7
Very Large Datasets

• Numerical simulation produces GBs and TBs of time
  series data
• How can we visualize this interactively on a 2GB
  workstation?
• Key ideas
• overview then focus (the visualization mantra)
• know the error in the data
• only read what you need

8
Data Management Components

• Granite Scientific Database System (Java)
• General support for rectilinear, multisource,
  multidimensional, multiresolution data
• I/O optimization based on iteration-aware
  prefetching and caching
• STARgen multiresolution data generation tool
• Batch-oriented general purpose command line tool
  that allows user defined output organization
  (C)
• STARgui limits output flexibility easy to use
  (Java)

9
Space/Time Wavelets

• Spatial wavelet transform applied to data from
  each step of time series
• Temporal wavelet transform applied to all data at
  corresponding positions in all steps

10
Quality of MR and AR data

• Scientists do not like discarding data
• Integration of error with the data is key
• Uncertainty visualization informs scientist
• Only delete time steps not significantly
  different from surrounding steps (based on d)
• Only abstract spatial regions with low error
• Tradeoff is that we can handle larger data
  interactively

11
Recent Accomplishments

• Significantly expanded error model
• multiple error measures at different resolutions
• Developed an interactive program to generate
  multiresolution data hierarchy of intermixed
  spatial and temporal resolutions.
• Interface to the VisIt environment
• will focus on this

12
Delivery Mechanism

• Multiresolution data management is our gig
• To be useful to our colleagues, we need a way to
  deliver this functionality to them
• We dont have the resources to build all the
  visualization tools they might ever want
• They use the VisIt package from LLNL
• We integrated our multiresolution data management
  model into VisIt

13
VisIt Functionality

• Visualization and analysis tool for 2D and 3D
  time series data
• scalar, vector and tensor field visualization
  modules
• multiple mesh types rectilinear, curvilinear,
  unstructured
• Elaborate gui for interactive viewing/data
  manipulation
• parallel/distributed architecture

14
VisIt Architecture

• Principal components
• Database interfaces
• Plots data renderers
• Operators can be applied to data prior to plots
• To facilitate distributed computing, there is a
  clean break between
• Viewer usually on local desktop
• Data engine can be on remote machines
• Dynamically loaded plugins

15
STAR / VisIt Interface

• STAR database plugin
• Accesses STAR multiresolution data hierarchy
• STAR operator plugin
• User controls resolution via an operator plugin
• Interaction with operator plugin triggers data
  reload

16
STAR/VisIt MR Support

• VisIt state after a STAR data object opened
• 1 slice of one high resolution time step shows
  density variate
• STAR operator dialog to control data resolution

Any VisIt rendering can be applied to any
compatible STAR data
17
STAR/VisIt MR Support 2

• Medium resolution
• Low resolution

18
STAR Error Data

• STAR error data is generated at same resolution
  as the lower resolution data
• Top is error of resolution 2 and bottom is
  resolution 2 data.

19
STAR/Visit Error Data

• Error is just another data set to VisIt top is
  error data drawn with opacity at 50
  superimposed on the medium resolution data,
• VisIt lets you drag a slider to to change opacity
  dynamically or swap views between the error and
  data.

20
VisIt Limitations

• VisIt is powerful and extensible BUT
• VisIt also has a key rigid architectural
  constraint
• its visualization is based on vtk, which assumes
  that all data needed for rendering will be in
  main memory throughout rendering
• VisIt makes the same assumption in particular
• it provides no way that a database plugin can
  update the data based on viewing parameters (for
  example)

21
Effect of VisIt Limitations

• We cannot change our resolution automatically as
  user zooms in or out
• wed like higher resolution data as range of
  interest gets smaller, lower resolution as it
  gets bigger
• We cannot implement out-of-core visualization
• We cannot utilize our iteration-aware prefetching
  and caching support
• BUT, half a loaf is better than none
• our collaborators have MR data access in VisIt

22
Conclusions

• Principal goal combine space and time
  multiresolution into unified data model
• Focus on simulation of MHD phenomena
• Integrate error model into application
• Integrate multiresolution support with error into
  the VisIt environment
• gives immediate access to a wealth of
  visualization and analysis functionality

23
Future Work

• Extend AR data support to spatial domain
• Extend support for multiple error representations
• Extend uncertainty visualization tools
• Investigate changes to VisIt internals

24
Acknowledgements

• Early aspects of this work was supported by the
  National Science Foundation
• Grants IIS-0082577 and IIS-9871859
• Principal collaborators Andrew Foulks (CS), Ted
  M. Sparr (CS), Xuan Tang (CS), Philip J. Rhodes
  (CS), John McHugh (ME)
• Current development is supported by NASA
• Applied Information Systems Research program
• Principal collaborators Andrew Foulks (CS), Ted
  M. Sparr (CS), Jimmy Raeder (Physics)
• Students David Benedetto, Sam Vohr
• Web site http//cs.unh.edu/star
• downloadable packages for STAR data generation
  and VisIt plugins

25
http//cs.unh.edu/star