180 likes | 360 Views
Sam Uselton Center for Applied Scientific Computing Lawrence Livermore National Lab October 25, 2001. Challenges for Remote Visualization: Remote Viz Is Really Large Data Viz. Remote Viz == Large Data Viz. The Real Problem is TIME, not Distance.
E N D
Sam UseltonCenter for Applied Scientific ComputingLawrence Livermore National LabOctober 25, 2001 Challenges for Remote Visualization: Remote Viz Is Really Large Data Viz
Remote Viz == Large Data Viz • The Real Problem is TIME, not Distance. • Large : Defined Relative to Available Resources • Data Size vs Bandwidth • Data Size vs Memory • Data Size vs Storage • Examples of “TOO MUCH” • 56Kb Modem vs 10’s of MegaBytes • 10Mb EtherNet vs GigaBytes • Gigabit EtherNet vs 100’s of GigaBytes
Some Issues are NOT Visualization Specific • How are Remote Sites Accessed? • Find Relevant Data? Same. • Demonstrate Authorization? Same. • Access Content? Same? • Can Security Be Guaranteed? • Same Security Requirements? • Implementation Issues?
Visualization Activity : A Model • Get Data • Find • Demonstrate Authorization • Select / Extract / Derive Data • Describe Visualization • Mapping to Geometric, Visual and Other Attributes • Scene, Viewing and Rendering Attributes • Generate Images
Visualization Activity : A Model • Get Data • Find • Demonstrate Authorization • Select / Extract / Derive Data • Describe Visualization • Mapping to Geometric, Visual and Other Attributes • Scene, Viewing and Rendering Attributes • Generate Images • Interaction: • Mapping Controls to Dynamic Attributes • Manipulate Controls
Remote Exploration is Harder Than Remote Presentation • Exploration Requires Interactive Choices • Interactions Affected By Latency (AND Bandwidth) • Time (of course) • Consistency (!) • Multiple Times • Variability of Impact • By Interaction Mode: • Haptic Head Tracking Hand Tracking • Indirect Manipulation Command Line • By Individual and Expectation
Direct Approach: Fixed Bandwidth Requirement (Good) High Bandwidth Requirement (BAD) MegaPixel Workstation 1M pixels x 3 Bytes x 60 hz = 180 MB / sec IBM T220 High Resolution LCD (or a Tiled Display) 9M pixels x 3 Bytes x 30 hz = 810 MB /sec Large Tiled Displays too. Alternatives: Distribute Images
Alternatives: Distribute Images … Cleverly • Smaller Windows • or lower resolution • Generic Compression (example) • Processing Overhead at BOTH ENDS
Alternatives: Distribute Images … Cleverly • Application Specific Compression (examples) • Better Compression (Sometimes) • Less Overhead (Sometimes) • Batch Mode: Make Movies, ftp, then Play Locally • OR … Make CDs and Ship
Alternatives: Distribute the Data • Large Data Means Long Delay • Increasing Chances of Failures • Large Data May Exceed Local Resources • Memory, Storage, … • … and Wasteful When Some (Most?) Data Is Not Used • Batch Mode: Make TarBalls, ftp, then Play Locally • OR … Make CDs and Ship
Alternatives: Distribute Graphics Information • Geometry, Colors, Textures, … • Local Control of View • Solves Latency Problem for Viewing Changes • Render Using Local Hardware • Fast and Cheap • Appropriate for Local Display • MAY Use Less Total Bandwidth, But Slower Starting
Alternatives: Distribute Geometry and App Data • Local Control of View • Local Color Mapping ... • Local Quantitative Querying ... • BUT MORE DATA - Impacting Both Time and Storage
Alternatives: Distribute SOME Geometry • View Dependence • View Culling • Level-of-Detail • Occlusion Culling • Progressive • Interruptable
Alternatives: Distribute Some DATA • View Dependent & Progressive • Trickier: Some Sort-First Processing • Extract Geometry Locally • Lower Latency for Changing Geometry (Good) • Heavier Processing Load at Lighter Resource (Bad) • Interruptable
Alternatives: What Works Best ? • It Depends !! • Time varying data, • Data ”Over There" vs Data ”All Around Me" • Dynamic View vs • Dynamic Parameter Mapping vs • Dynamic Geometry Selection • Systems Should Support Multiple Alternatives
Comments On “Immersion” • Dynamic Head Tracking Controlling View of Scene • Powerful Qualitative Impact on Viewer (Good) • Stringent Latency Demands, Double Images (BAD) • Very Useful for Training and Planning • Less Important for Analytical Tasks
Comments On Collaboration • Group Activity Models: • Presenter(s) and Audience • Simultaneous Independent Activities • Tightly Coordinated Joint Tasks • Asynchronous Activities • Which Modes Are Needed For Particular Uses? • How to Move Between Models ? • How to Decide ? • How to Indicate ?
Acknowledgements • David Metz and KGO-TV for Video of the 2001 San Francisco Grand Prix Bicycle Race. • ASCI VIEWS, especially the LLNL visualization team. • This work was performed under the auspices of the U.S. Department of Energy by University of California Lawrence Livermore National Laboratory under contract No. W-7405-Eng-48. UCRL - PRES-144889