Graphics Systems and Toolkits

by Darrell and Mike

Key Considerations for Large Display Graphics

• Performance

• Portability

• Scalability

• Cost

• Bandwidth

Architectures for Distributed Graphics

• Client-server

  • Pros

    • Can achieve excellent performance in some cases

  • Cons

    • High bandwidth requirements

• Master-slave

  • Pros

    • Low bandwidth

    • Performance should be equivalent to single display

  • Cons

    • Synchronization can be difficult

Existing Tools for Distributed Graphics

• Chromium

  • Pros

    • Extreme portability.  Can theoretically run any OpenGL application.

    • Extreme flexibility.  Can write custom stream processing units (SPUs) to maximize performance.

    • Fast startup for applications

  • Cons

    • Performance issues as display size increases

    • Writing SPUs to maximize performance is difficult and sacrifices portability

• OpenSG

  • Pros

    • Built directly on top of OpenGL

    • Can be run on any system with OpenSG libraries installed

    • Excellent performance

  • Cons

    • Must learn a new API

    • Only useful if you need to use a scene graph

    • Slow startup

• VRJuggler

  • Pros

    • Can use different underlying graphics toolkits

    • Good performance, depending on the choice of toolkit

    • Input framework included

  • Cons

    • Synchronization issues (master-slave architecture)

    • Slow startup

• SAGE

  • Pros

    • Good performance (?)

    • Reasonable portability

    • Reasonable scalability

    • Multi-user

  • Cons

    • Not much has been implemented yet

• Graphics Supercomputers (e.g. SGI Onyx)

  • Pros

    • Excellent performance

    • Relatively easy to program (no networking/synchronization issues)

    • Can use sort-middle approach

    • Good portability

    • Good scalability

  • Cons

    • Very, very expensive

Performance vs. Portability

In general, for a tiled display, the higher the performance you desire, the more portability you must sacrifice, and vice versa.

Bandwidth

This will be the major bottleneck as display size increases.  It is estimated it will take as much as 45 years for computer-to-display wiring to achieve the bandwidth necessary to support a gigapixel display.  Is it time for a different rendering paradigm (e.g. adaptive frameless rendering)?

Parallel vs. serial

Where is the pipeline divided?

Sort-First

Sort-Last

Sort-Middle (Supercomputers only!)

Hardware Requirements

Cluster

Single Machine

Shared-memory