Presentations at Workshop 7

Computational Issues

• Sociology.
  • Main-frame decentralization to workstations
  • Set unconscious limits on scale of problems attempted.
  • Eased pressure for migration tools to parallel computing.
  • This initiative must dramatically reverse these trends.
• Scalable sparse matrix operations.
  • Locality of interactions yields irregularly sparse matrices.
  • Matrix-matrix operations often dominate iterative methods.
• Visualization. For massively parallel codes,
  • Large-data-base codes on remote machines demand real-time visualization tools, sharable by distributed collaborators.
  • Real-time performance and debugging requires highlighting of unusual performance or deviation from expected behavior.
  • Real-time visualization requires rendering on massive parallel architecture to deliver fast serial visualization by local user.
• Remote steering.
  • Remote users need to interactively steer simulation in space/time.
  • Software must ensure each viewer has time-coherent view.
  • Software must ensure coherency across all multiple views.
  • Simulation must be robust with respect to all such inspections.
  • Application software must support variety of local software.
• Integrated development environment. Any given user
  • Can simply construct interface for particular application ignorant of programming paradigm or visual environment details;
  • Can see view of overall flow of program;
  • Click to see I/O between two routines;
  • Manipulate subalgorithms to construct/adapt code;
  • Access a new subprocess within context of entire code.