Framework | Docker & Nvidia-Docker

Introduction

1.1 tips

• a container is designed to virtualize a single application, and all containers deployed on a host share a single OS kernel.¹
• docker containers are platform-agnostic, but also hardware-agnostic.¹
• docker images are built layer-by-layer and the layers can be shared with multiple images to save disk space on the host (as well as deployment time).¹
• there are some differences when installing nvidia-docker between version 1.0 and version 2.0, nvidia-docker 1.0 must register runtime, just follow as nvidia-docker github. More detail can be seen in this 优快云 blog.²

Installation

1.1 download

• download the docker from link (password: 6i8g)

1.2 install docker-ce and nvidia-docker 2.0

# cd docker foler

sudo dpkg -i libltdl7_2.4.6-2_amd64.deb
sudo dpkg -i docker-ce-cli_19.03.3~3-0~ubuntu-bionic_amd64.deb
sudo dpkg -i containerd.io_1.2.6-3_amd64.deb
sudo dpkg -i docker-ce_19.03.3\~3-0\~ubuntu-bionic_amd64.deb
docker -v

# nvidia-docker 2.0
sudo dpkg -i libnvidia* nvidia*
sudo pkill -SIGHUP dockerd
nvidia-docker -v 

1.3 alter to ali source

sudo gedit  /etc/docker/daemon.json

# append
{
    "registry-mirrors":["http://hub-mirror.c.163.com"] 
}

# restart docker
sudo systemctl daemon-reload  
sudo systemctl restart docker

1.4 add user to docker group

sudo gpasswd -a ${USER} docker

sudo systemctl restart docker

newgrp - docker

Reference

---

1. Nvidia Docker ↩︎ ↩︎ ↩︎

2. 优快云 blog ↩︎