Contents

• Contents
• Introduction
  • The need for parallelization
  • A computing platform
  • The TONS idea
  • ``Interpreted'' code
  • Interpreted code and performance
• Designing the system
  • System overview
  • The user's view
  • The scheduler
    • Sub-clustering
    • The parallelization engine
    • The scheduler logic
    • Keeping everyone busy
  • The node server
• The language
  • Existing languages
    • Some problems with C and Fortran
  • The new language
    • Data types
    • Side effects
    • Introduction to the VM langauge
    • High-level opcodes
    • Function calling
    • Language constructs
    • Indexing
  • CSP and job distribution
    • CSP and the virtual machine
    • Parallelization paradigms
• Parallelization
  • Properties of parallelization
  • Communication and ``tasks''
  • Sequence parallelization
    • In case we have no parallel tasks currently
    • In case we already have parallel tasks
    • Overview
  • Loop parallelization
    • Identifying common instructions
    • A quick heuristic
    • Going parallel
    • Creating the loops
    • Fitting in the common instructions
    • Overview
  • Instruction expansion
  • Nested loops
  • Where to go next
• Performance
  • VM performance
  • The Fibonacci test
  • Matrix multiplication
• Conclusion
• Benchmarks
  • The Fibonacci calculator
    • The C implementation
    • The Perl implementation
    • The MatLab implementation
    • The TONS implementation
  • Matrix Multiplication
    • The MatLab implementation
    • The TONS implementation
• Bibliography

<