Skip to content

Latest commit

 

History

History
61 lines (42 loc) · 2.84 KB

specification.md

File metadata and controls

61 lines (42 loc) · 2.84 KB
Raw

Edge Appliance Specification

[TOC]

Reference Hardware Design

HW Block Diagram

image

With our hardware design, there are simple local clusters in the appliance, all device nodes are connected via Ethernet, and use internal gateway for internet access.

  • Internal network connection linked with 1Gbps Ethernet
    • PoE switch enable PoE perhaps devices connectives, such as IP Camera, etc.
    • Static network topology inside the appliance
  • OOBM Controller provide physical compute node control
    • Power/Reset/Recovery signal
    • Serial console access
    • Optional power wire control
  • GPU Compute Node up to 24 nodes
    • Starts from minimal 3 nodes

Reference Hardware

image

In our sample hardware, we choose NVIDIA Jetson AGX Xavier and Jetson TX2 as our compute nodes. The reason for those chosen is we're looking for GPU enabled low-cost devices for both GPU related computing and general computing purpose. Developers are able to use any other hardware that meets all requirements for this software. This selection guideline are also applied for OOBM control board: Raspberry Pi 4. Developers can change control board to any board that supports linux OS running plus ethernet & GPIO pins for compute node device control.

In this reference hardware design, we choose relay board for traditional electrical isolation, but we also recommend that any production-ready environment should use more reliable devices instead of relay, e.g., optronics relay or CMOS switch circuit.

Software Architecture

image

Application Orchestration

The sample software uses native kubernetes as orchestration engine, we also use it to manage system-wide services for edge maintenance tasks.

Additional software stack achieved our goal of the edge appliance:

  • Easy Configuration
  • Easy Deployment
  • Federation support for multiple appliances applications
  • Internal HA & Failover for applications

Those features are explained in detail under their sub-projects.

Deployment

Scalable Appliance Configuration

image

  • Working like as standalone GPU enabled clusters
    • Suitable for microservices applications
  • Configurable nodes for scalable application uses

Cloud Service Integration

img

  • Use appliance as edge devices, integrated with cloud providers for edge computing applications
    • Integrated with cloud management
    • Support multiple appliances deployment
    • Suitable for AI based edge use cases