slide 2 of 37

Architectural Model

• Multiple independent machines
• Processor and total I/O performance both scale
• Each interprocessor communication is an OS call
  • High latency
  • Bandwidth best for larger block transmissions
• Can sometimes overlap compute and communication