Skip to content

Latest commit

 

History

History
128 lines (93 loc) · 5.12 KB

docker_for_aws.md

File metadata and controls

128 lines (93 loc) · 5.12 KB
Raw

Docker For AWS

AWS Credentials

Prior to following this tutorial, you will need to have configured your AWS command line interface with the proper credentials.

aws Command Line Interface

You will also need to have the aws command line interface installed. You can do that using pip by following these instructions.

GPU on AWS

The general workflow herein supported consists of the following:

  1. Do development work on your project locally using the standard docker image
  2. Sync this work with your github account
  3. Start a new GPU instance on AWS using docker-machine with its built-in EC2 driver
  4. SSH into the GPU instance using docker-machine
  5. Use nvidia-docker tool to run the GPU configured docker image

Local Development & Sync with Github

Note that this means sync to your personal github account.

  1. Develop your work locally and manage it using git
    • develop your work locally using your standard workflow (python modules/scripts, jupyter notebook)
    • git add -A && git commit -m 'Standard commit message'
  2. Push local changes to github
    • git push

Start a new GPU instance

WARNING: This is the phase at which you will start to accrue costs. Running this command will start a GPU instance on AWS.

To manage your GPU instance, you will be using three basic commands:

  1. docker-machine create
    • creates a new docker host machine
  2. docker-machine ls
    • lists all running docker host machines
  3. docker-machine rm
    • removes a docker host machine

Make sure not to leave your GPU instance running. Always use docker-machine rm to destroy your instance when you are not using it.

  1. Configure the jupyter security group.

  2. Create an EC2 GPU instance.

    docker-machine create --driver amazonec2 \
                      --amazonec2-region us-west-2 \
                      --amazonec2-ami ami-a9d276c9 \
                      --amazonec2-security-group jupyter \
                      --amazonec2-instance-type g2.2xlarge \
                      udacity-carnd-gpu

    You may run into the error:

    Error with pre-create check: "There is already a keypair with the name 
udacity-carnd-gpu. Please either remove that keypair or use a different 
machine name."

    This is an error with AWS. You can either change the name you wish to use or navigate to EC2 > Network & Security > Key Pairs and remove the key pair with the conflicting name.

  3. Provision your new EC2 GPU instance for running nvidia-docker

    If you create a custom AMI, you can skip steps 2 and 3 in the future.

    1. SSH into your instance

      (LOCAL) $ docker-machine ssh udacity-carnd-GPU

    2. Configure your instance for nvidia-docker

      (EC2-GPU) $ sudo apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub
(EC2-GPU) $ sudo sh -c 'echo "deb http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64 /" > /etc/apt/sources.list.d/cuda.list'
(EC2-GPU) $ sudo apt-get update && sudo apt-get install -y --no-install-recommends cuda-drivers

    3. Install nvidia-docker

    4. Run the test image

      (EC2-GPU) $ sudo nvidia-docker run --rm nvidia/cuda nvidia-smi

      After the image is finished pulling, you should see the following:

      Status: Downloaded newer image for nvidia/cuda:latest
Wed Dec 28 00:04:07 2016
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 367.57                 Driver Version: 367.57                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GRID K520           Off  | 0000:00:03.0     Off |                  N/A |
| N/A   29C    P8    18W / 125W |      0MiB /  4036MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

  4. Use nvidia-docker to launch a new carnd image with GPU support.