Embedded System Project

Table of Contents

1. Power-Performance Characterisation analysis
2. Power-Performance Report
3. Governor Development
4. Governor Optimization
5. Governor Report

Introduction

Embedded systems operate in power-constrained environments, and governors are essential for power management in these systems. This project focuses on developing a new embedded system Governor from scratch, specifically designed for power management in a Machine Learning (ML) application.

System Requirements

• Laptop running Ubuntu 20.04 with a USB-A port (other OS not supported in this lab)
• USB-A adapter required if the laptop lacks a built-in USB-A port

About Governors

Governors are Operating System (OS) sub-routines responsible for managing power in embedded systems [3]. They can be either general-purpose or application-specific. We have created an application-specific Governor to perform power management for a Machine Learning (ML) application.

ML Application

We will focus on power management for a Convolutional Neural Network (CNN) inference, a popular ML application. Five state-of-the-art image classification CNNs will be used: AlexNet, GoogleNet, MobileNet, ResNet50, and SqueezeNet. These CNNs will be executed using the ARM-CL framework on ARM-based Heterogeneous Multi-Processor Systems on Chips (HMPSoCs).

About HMPSoCs

HMPSoCs are Integrated Circuits (ICs) that integrate different processing cores on a single die. They are widely used in portable embedded devices like smartphones and gaming handhelds. In this project, we will work with the Amlogic A311D HMPSoC in the Khadas Vim 3 embedded platform.

Project Structure

The project consists of the following components:

1. Power-Performance Characterisation Analysis: This section analyzes the power-performance characteristics of the embedded system.
2. Power-Performance Report: A report summarizing the findings and insights from the power-performance analysis.
3. Governor Development: Where we will write the new embedded system Governor.
4. Governor Optimization: This section involves optimizing the Governor's performance and power management capabilities.
5. Governor Report: A detailed report showcasing the Governor's development, optimization, and performance.

Getting Started

To run this project, make sure you have the necessary hardware and software requirements and clone this repository to your local machine.

Note

This project is designed for laptops running Ubuntu 20.04 with a USB-A port.