Edge Deployment Technology Selection Criteria

  1. Use Case – Why Edge?:
    • Custom model deployments v. End to end solution, e.g.: Instore security, OH&S Systems etc.
    • Faster inferencing – closer to source?
    • Data residency?
    • Model should be able to run if connection to Azure is lost?
    • Target site has limited to zero internet connectivity?
  2. Target compute type (also informed by customer skills maturity):
    • VM -> IOT Edge? ARC for VMs?
    • K8s - > Arc for K8s?
  3. HA and DR requirements for edge models:
    • Criteria? SLAs etc.
  4. Azure Parity requirements:
    • AAD / AD integration?
    • Storage or other PaaS support
  5. Target compute spec requirements:
    • CPU, GPU, FPGA?
    • Infrastructure is custom or MS edge offerings like Stack Edge, Stack Hub, HCI
    • Custom hardware like Percept
  6. Support for specific model frameworks:
    • Intel LVA? Nvidia Triton? Onnx? etc.