Edge Computing

(Master of Information Science) Lecture on IoT and Edge Computing

• It is difficult to do all the calculations by communicating with the server.
• Therefore, the approach is to perform calculations on the edge.
  • In the limited energy and computing resources, we strive for approximation calculations, etc.
• What is required for edge computers?
  • Real-time processing
  • Generality and flexibility
  • Energy efficiency
  • 1 is a prerequisite
  • On top of that, 2 and 3 are conflicting challenges#trade-off
    • If we create a CPU for generality, it consumes energy.
    • If we create a processor dedicated to a specific application for energy efficiency, it lacks generality.
  • What to do?
    • Method 1: CPU + accelerator
      • ASIP: A combination of a general-purpose processor and an application-specific accelerator.
        • The accelerator handles particularly heavy specific processing.
        • “The best of ASIC and GPP.”
      • Application non-specific accelerator
        • Aims for higher generality and energy efficiency than ASIP.
        • Example of use? Approximate calculation in image processing (approximate calculation based on calculation reuse)
    • Method 2: Domain Specific Architecture