[Next] [Previous] [Top]

Strategic Overview

Strategic Overview

Why Parallel Computing?

• Need for speed?
  • time critical modelling problems such as the weather forecast - forecast for tomorrow must be completed in 5 hours

• Problem too large? - Teraflop requirements of "Grand Challenge" research (109 floating point operations per second)

• Nearing the physical limits for single processor technology?

• Every major computing manufacturer is developing a parallel architecture machine

• Large databases: search times; access times; data mining

Strategic Overview

Pros

• Cost performance ratio (Mflops/$)

• Performance possiblities
  • peak

  • ultimate

  • actual

  • scalable

  • affordable

  • high reliability - fault tolerance

Cons

• Porting existing Software
  • restructure (not maintenance)

  • inadequate development environments

  • specialized training

• Technology traps
  • processors are developing rapidly

  • communication speeds are lagging behind

  • systems software further behind

  • application software even further behind

• Consequences
  • long development cycle

  • upgrade problems

Summary

• Parallelism is essential

• Avoid leading edge technology - use tried and tested technology

• Do not expect miracles

• Adopt parallel programming ideas as soon as possible