19
MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking

MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Embed Size (px)

Citation preview

Page 1: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

MEAD: Volume VisualizationDavid Porter, U. Minnesota/LCSE

•Data Pipeline

•A3D

•HVR

•Tiled Commas Output

•Windows & Unix

•Formats & Frameworks linking to MEAD

Page 2: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Data PipelineHydro-code:WRF/COMMAS

ROMSRaw Fields (U,V,W,…)Formats: (HDF5, WRF, cdump)

Data Post-Processing:A3D

Block-Tree HierarchiesDerived Fields (Humidity, Vorticity)

Volume Visualization:HVR

Raster Data: (Images, Movies)

Page 3: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

A3D: More Than Just Compression, Conversion, Scaling

•Reads in raw 3D fields onto a stackTiled cdump, BoF, HDF4, HDF5, netCDF?

•Performs calculations on this stack fields+-*/, d/dx, spatial filters, map from sigma-coords. (mm5)

•Produces statistical reductionspower spectra, structure functions,histograms, radial & vertical profilesBoF, BoB, SoF, SoB files cross-correlations

as well as Block Tree hierarchies for visualization.

Energy Spectra(k) MLT coefficients(Z) MM5 simulation

Page 4: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Block Tree Hierarchy: Basic Idea

Illustration: A 2 level hierarchy in 2D

• 16x16 array is split into 4 tiles•Each tile is sampled at a stride of 2x2•8x8 array of info in each node of tree

Page 5: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

3D Block-Tree •A 3D mipmap

•Browse even the largest data sets at interactive speeds.

•Full size of data only limited by disk.

•Trees can be both logically and spatially irregular.

•Highest resolution need not span entire domain.

•Block size tuned to both hardware and application. = 128x128x64

256

32

4

1

Page 6: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

HVR•Organize voxel data into a Block-Tree hierarchy

Stage large volumes of data on diskData compression: 1 byte per voxel & Tree PruningControl rendering quality vs. speed via tree traversalDCOM rendering server: HVR-SERVER

•Use key framing to specify complex path in space and timeInteractive GUI driven: HVR-GUI

•Tree hierarchy + key framing facilitate remote renderingRoot of Block-Tree hierarchy ideal for local key framing needsKey framing software delineates division of labor between

Local and remote resources & work.Optimized use of network bandwidth

Page 7: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Rendering Tolerances

Tolerances based onAngle subtended by voxelColor map index errorAbsolute resolution

(For speed only)

2D illustrationQuad-tree hierarchyEach tile might be 256x256

Page 8: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Tolerance: Based on Angle Subtended

Viewpoint

If a voxel at a given resolution subtends to great an angle, then scan down to finer resolution.

Page 9: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Tolerance: Based on Max. Error in Value

•Nodes have bound on error

•Const color no error

•Stop at nodes w/ no error

Page 10: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Remote Rendering: Routines

Local

•HVR-GUI•HVR-SERVER•MOVIE PLAYER

Remote

•HVR-SERVER

Page 11: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Remote Rendering: Data

Local

•Finder Bricks•Key Frame Parameters•Color & Alpha LUT•Low-Mid Sized Movies

Remote

•Block-Tree Hierarchy•Key Frame Parameters•Color & Alpha LUT•Low-Mid Sized Movies •Power Wall Movies

Tiny KB-Small MBMedium GBLarge TB+

Page 12: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Example: Tiled COMMAS•PC-cluster simulation produces “tiled output”

Titan clusterLogistics: 103 files randomly distributed across ~102 nodes

•Tiles are collected to post-processing hostsSCP to LCSE PC-rendering clusterLogistics: cdump files redistributed across 10-100 nodes

•A3D reads tiled cdumps and produces derived scalar volume data in block tree format.

Fields: W, vorticity, cloud water (*.hv files)Logistics: hv files redistributed to rendering cluster

•HVR produces images & movies

Logistics: Need data a base: MySQL

Page 13: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

COMMAS: 96x96x36 (4x4 tiles)

Page 14: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

COMMAS: 250x250x72 (5x5 tiles)

Page 15: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Porting to Unix

Windows Unix

OpenGL OpenGLWGL GLUTDCOM CORBA? Condor?

GUI: MFC FLTK? XML? Command line?Subroutine? Your framework?

Page 16: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Data Formats

I’m looking for a few good data sets.•WRF•ROMS

In a few relevant formats•netCDF•HDF5

Data Sets Please …

Page 17: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Frameworks

I’m looking for a few good frameworks

Connecting a few relevant centers

APIs Please …

Page 18: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

Web Link

http://www.lcse.umn.edu

•Downloads of volume rendering apps for your PC

•Downloads of software under GNU GPL

•Many volume rendered movies

•User Guide to the Graphical user interface: hvr_gui

•Description of the Hierarchical Volume Rendering server hvr_server

•Software for producing volume data in Block Tree format

Page 19: MEAD: Volume Visualization David Porter, U. Minnesota/LCSE Data Pipeline A3D HVR Tiled Commas Output Windows & Unix Formats & Frameworks linking to MEAD

LCSE MEAD Milestones

(Q1) Simplified code (COMMAS) running w/ SHMODHVR Vol Vis of WRF datasets via data postproc w/ a3d

(Q2) Simplified code with dynamic load balancingHVR vis of ROMS data.

(Q3) Simplified code running in fault tolerant mode

(Q4) Simplified code running on TeraGridHVR vis of data on TeraGrid.