Docker Example Application

• What is this repository for?
• Terminology
• How to get set up?
• Install and Register Nvidia
• Testing Docker in this Repo
• Who do I talk to?

What is this repository for?

• To show how to use docker with an example project

Terminology

• Dev Machine - the device where the application development is being done (most likely a laptop)
• Target Machine - the device that the containerized image is meant to be run on.
• Hosted Machine - the device that holds the docker container. Can be the dev or target machine.

How to get set up?

• For Ubuntu 18.04, install docker engine from here
• For additional help on installing docker for Ubuntu 18.04, see docs/DockerSetup.md
• For other OS, install docker desktop for GUI interface. Download and setup found here
• Create a docker hub account here.
• To be able to push to docker hub, delete the file config.json inside ~/.docker on the hosted machine (verified on Ubuntu 18.04)
• If Nvidia CUDA support is needed, follow directions in section Install and Register Nvidia. This has not been tested.
• The example in this repo is based on this link
• Source for creating images for specific architecture systems found here
• Source for cross compiling (much faster) found here
• Specific project work flow can be found in docs/Workflow.md

Install and Register Nvidia

• It is uncertain if CUDA support is needed, but this sections shows how to set it up if needed.
• This section is only necessary if CUDA support for the image is needed.
• Do this before creating the image if CUDA support is necessary.
• The following steps are implemented on the host machine that is creating the image.
• It is not tested of verified if this needs done on the target device as well.
• Link to installing Nvidia drivers for different systems can be found here
• To install Nvidia drivers in Ubuntu 18.04, run the following

sudo apt-get install linux-headers-$(uname -r)
distribution=$(. /etc/os-release;echo $ID$VERSION_ID | sed -e 's/\.//g')
wget https://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64/cuda-keyring_1.0-1_all.deb
sudo dpkg -i cuda-keyring_1.0-1_all.deb

• See post install actions for mandatory setup
• Registering Nvidia and options link can be found here
• To register Nvidia with docker, run the following on the hosted machine.

sudo dockerd --add-runtime=nvidia=/usr/bin/nvidia-container-runtime [...]

• Specify docker cuda capabilities that are desired for the image when running on the hosted machine.
• See two separate examples below that can potentially be used:

docker run --rm --runtime=nvidia \
    -e NVIDIA_VISIBLE_DEVICES=2,3 \
    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
    nvidia/cuda nvidia-smi
    
docker run --rm --gpus 'all,"capabilities=compute,utility"' \
    nvidia/cuda:11.0.3-base-ubuntu20.04 nvidia-smi

• User can avoid calling out capabilities and options when running the docker image.
• In DockerFileBuildEnv, add the following to set up cuda capabilities without needing to specify with command line at runtime.

ENV NVIDIA_REQUIRE_CUDA "cuda>=11.0 driver>=450" # restricts cuda version and driver
ENV NVIDIA_VISIBLE_DEVICES all
ENV NVIDIA_DRIVER_CAPABILITIES compute,utility

Testing Docker in this Repo

• This section verifies docker is installed and working properly on the host machine.
• The test was created for x86_64 Linux Ubuntu 18.04 host machine. Later versions of Ubuntu may work, but are not verified.
• An image is needed to run this test.
• There are pre-made images available to pull.
• To get the hello world pre-made linux/amd64 image, run:

sudo docker login
sudo docker pull jakeinit/hello_world:0.00.01

• If an image needs to be created, see docs/CreateImage.md
• For the hello_world image, make sure these parameters are set in the top level make file:

1. PROJECT_NAME=hello_world
2. DOCKER_DEPS_VERSION=0.00.01
3. DOCKER_HUB_USER=jakeinit
4. DOCKER_SYS_ARCH=linux/amd64
5. HOST_MOUNT_PATH=${CURDIR}

Then run:

sudo make login

• The above will open the image in a folder named "hello_world" with current host directory files and folders mounted inside.
• All changes done to files and folders in image or on host machine will be reflected on both image and host.
• exit the image with "ctrl + d"
• Run

sudo make build

• This will build the hello world binary in the image architecture format and will reside in the host machine build folder
• The binary can be tested with

./build/Example

• Output: "Hello From Docker"

Who do I talk to?

• Jacob Morgan [email protected]