README_en.md

DAPart

This open source code is a real test platform built by the actual experiment in the paper "DAPart: An Online DRL-based Adaptive Partition Framework for DNN Models in Edge Computing".

DAPart

Explore the documentation for this project »

View Demo · 中文 · English

This README.md is for developers.

Catalogue

• Getting started guide
  • Configuration requirements before development
    • Client devices
    • Server device
  • Installation procedure
• File directory description
• Deployment and operation
• Contributors
• Version control
• Author

Getting started guide

The open source code is divided into two parts: the client device uses Jetson Nano, and the server device uses a computer with a Linux system. The experimental equipment in this paper is shown in the table below.

Hardware	User Equipment Device (Jetson Nano)	Edge Server
System	Ubuntu 18.04.6 LTS	Ubuntu 22.04.2 LTS
CPU	4-core ARM [email protected]	Intel(R) Core(TM) [email protected]
GPU	128-core Maxwell@921MHz	GeForce GTX 3090 24GB
Memory	4GB LPDDR4 25.6GB/s	4*16GB LPDDR4 3200 MT/s
Hard Disk	64GB microSDXC 140M/s(max)	11T SSD + 42T HDD
Network Connection	WiFi 2.4G:300Mbps 5G:867Mbps	Ethernet 1000Mbps

Major environment configuration requirements before development

Client devices (Jetson Nano only allows this version of the environment)

1. python==3.6.15
2. torch==1.4.0
3. torchvision==0.5.0
4. tegrastats
5. jtop

Note: The Jetson Nano environment installation process is detailed in the official reference documentation

Server device

1. python>=3.7
2. torch==1.13.1
3. torchvision==0.13.1

Installation procedure

1. Clone the source code of the repository

git clone https://github.com/Jma512/DAPart.git

2. Installation environment Configure the necessary packages

File directory description

DAPart 
├── /data/
│  ├── /test/
│  │  └── ...
│  └── ...      //The images needed to simulate the task during the experiment
├── /model
│  ├── /mobilenetv2/
│  │  │  └── /logs
│  │  ├── downloadmobilenetv2.py
│  │  └── mobilenetv2_pretrained_imagenet.pth
│  ├── /resnet50/
│  │  │  └── /logs
│  │  ├── downloadresnet50.py
│  │  └── resnet50_pretrained_imagenet.pth
│  ├── /vgg16/
│  │  │  └── /logs
│  │  ├── downloadvgg16.py
│  │  └── vgg16_pretrained_imagenet.pth
├── DAPart_Edge_Server.py
├── DAPart_User_Equipment.py
├── experiment_neuro.py
├── mobilenetv2.py
├── resnet50.py
├── vgg16.py
└── README.md

Deployment and operation

The code can be deployed on the client side and the server side respectively, and the server side runs DAPart_Edge_Server.py, the client device runs DAPart_User_Equipment.py

Contributors

xxx@xxxx(Keep private)

Version control

The project uses Git for version management.

Author

xxx@xxxx(Keep private)