EMLI CONTAINER STACK - QSG

AS OF 9/14/21

Congratulations on purchasing a docker integrated system for Deep Learning. Below is a Quick Start Guide for Docker and Nvidia-docker.

Docker images available on this system:

nvidia/cuda
nvidia/caffe
nvidia/digits
portainer
tensorflow/tensorflow:latest-gpu
PyTorch
RapidsAI

Docker Command line Option:

To pull additional Docker image (from NGC Repository)

# Download / pull images for NGC Repository

root@u105724:~# docker pull nvcr.io/nvidia/cuda:9.1-devel
9.1-devel: Pulling from nvidia/cuda
976a760c94fc: Already exists
c58992f3c37b: Already exists
0ca0e5e7f12e: Already exists
f2a274cc00ca: Already exists
708a53113e13: Already exists
2ec2fca7a49c: Pull complete
34026c3e50ea: Pull complete
0e4a761cbcd3: Pull complete
2d1d54944b4e: Pull complete
Digest: sha256:5c91a161147220b06624cc490877b5b3867c13e86d5ee40d0e0fe6d5117f2137
Status: Downloaded newer image for nvcr.io/nvidia/cuda:9.1-devel
nvcr.io/nvidia/cuda:9.1-devel
root@u105724:~#

View pulled Images on system

# docker images to see installed images

root@u105724:~# docker images
REPOSITORY                        TAG                           IMAGE ID            CREATED            SIZE
nvcr.io/nvidia/cuda               9.2-devel                     1874839f75d5        7 weeks ago        2.35GB
nvcr.io/nvidia/cuda               10.0-devel                    f765411c4ae6        7 weeks ago        2.29GB
nvcr.io/nvidia/cuda               10.1-devel                    9e47e9dfcb9a        6 weeks ago        2.83GB
nvcr.io/nvidia/cuda               10.2-devel                    af2eaa345ab7        6 weeks ago        2.91GB
nvcr.io/nvidia/caffe              20.01-py3                     6094e9a70920        8 weeks ago        4.85GB
nvcr.io/nvidia/digits             20.01-tensorflow-py3          1430fdae6f40        7 weeks ago        9.36GB
nvcr.io/nvidia/tensorflow         20.01-tf1-py3                 e9f1a32f9cad        7 weeks ago        8.39GB
nvcr.io/nvidia/pytorch            20.01-py3                     5c0c8c90f238        7 weeks ago        9.12GB
nvcr.io/nvidia/rapidsai/rapidsai  cuda10.1-runtime-ubuntu18.04  7694f69b386c        5 weeks ago        8.87GB
portainer/portainer               latest                        ff4ee4caaa23        6 weeks ago        81.6MB
root@u105724:~#

View all containers on the system (including running and stopped)

# docker ps // to see all active containers/container info
# docker ps -a // to see all containers/container info

[root@c101086 ~]# docker ps -a
CONTAINER ID   IMAGE                COMMAND             CREATED             STATUS             PORTS                             NAMES
6a4d72fc7197   nvidia/digits        "python -m digits"  30 seconds ago      Up 29 seconds      0.0.0.0:5000->5000/tcp, 6006/tcp  digits-333201001-0
cdae95d22e84   portainer/portainer  "/portainer"        About a minute ago  Up About a minute  0.0.0.0:9000->9000/tcp            portainer-021237417-0
[root@c101086 ~]#

Run command inside of the container (interactively)

# to execute a shell within the container
[root@c101086 ~]# docker run --runtime=nvidia --rm -it nvidia/cuda bash
root@6c83ee4f8141:/#
# you will see hostname change to the container ID you are now in

For additional docker images, please go to: https://hub.docker.com/

NVIDIA Digits

DIGITS Quickstart Script (found in the root's home folder Directory and /usr/local/bin)

This is now also loaded in /usr/local/bin/startDigits so you may run #startDigits from anywhere to start a new unique container

[root@localhost ~]# cat startDigits.sh
#!/bin/bash
DATE=$( date +%N )

docker run --gpus all -it --name digits-$DATE-0 -d -p 5000:5000 -v /data/datasets:/opt/datasets --restart=always nvcr.io/nvidia/digits:20.01-tensorflow-py3

# Using /data/datasets on the host for Digits to access the data files
#options
# --runtime=nvidia, specific for passing the nvidia-docker
# -e NVIDIA_VISIBLE_DEVICES="0,1,2,3" control which Nvidia GPU to pass to the container
# --name = to name the container of your container
# -$DATE-0 variable implemented to create unique container names when starting a new one
# -d = detached process to run the container in the background
# -p = specify port (host port:container port)
# -v = volume, to link a directory from the host system to the container (host directory: container directory)
# --restart=always, set container to start after every restart
# nvidia/digits = specified docker image to load container
# in summary, this script will create a nvidia/digits based container using GPUs 0,1,2,3.  Container name = digits-$date-0 listening at port 5000 and linking the /home/data host filesystem to /opt/datasets within the container.
# you can access the web GUI via web browser // use <hostsystemIP>:5000

Portainer

Portainer is a simple management solution for Docker. Easily manage your Docker hosts and Docker Swarm clusters via Portainer web user interface.

[root@c101086 ~]# docker images | grep portainer
portainer/portainer latest 47dbf4321bb4 4 weeks ago 10.7MB
[root@c101086 ~]#

# to create a new Portainer Container
[root@localhost ~]# docker run -d -v "/var/run/docker.sock:/var/run/docker.sock" -p 9000:9000 portainer/portainer

# you can access the web GUI via web browser // use <hostsystemIP>:9000
# you will be prompted to enter an admin password, choose manage docker where portainer is running and connected

Initial portainer container instance password configured on the system is : password@1

If the portainer container was removed, then the end user will have to supply a new password for the new container instance.

Initial Startup / Configure for a new instance of Portainer:

Type in a password for admin

Click on Create User to continue.

Portainer allow different way to connect to docker engine, Select Local to manage local docker setup.

Click on Local and then click on Connect at the bottom.

Portainer will confirm the selection.

Click on local at the bottom of the screen to go to the dashboard.

Dashboard View -

provides an overview of the container(s) running on the systems, along with the related volumes and network info.

Portainer Containers view -

Overview of loaded containers status, and control / manage of the containers

Portainer Images View -

Overview of pulled images on the system, or download (pull) additional images available at the DockerHub Registry

Rapids Container and Notebook Server

NOTE: This will run JupyterLab on port 8888 on your host machine.

Command:

docker run --runtime=nvidia --rm -it -p 8888:8888 -p 8787:8787 -p 8786:8786 nvcr.io/nvidia/rapidsai/rapidsai:cuda10.1-runtime-ubuntu18.04
utils/start-jupyter.sh

[root@c105017 ~]# docker run --runtime=nvidia --rm -it -p 8888:8888 -p 8787:8787 -p 8786:8786 nvcr.io/nvidia/rapidsai/rapidsai:cuda10.1-runtime-ubuntu18.04
 
## Starting jupyter service
(rapids) root@712e75ae4a0e:/rapids/notebooks# bash utils/start-jupyter.sh
  
jupyter-lab --allow-root --ip=0.0.0.0 --no-browser --NotebookApp.token=''
  
[I 19:26:58.713 LabApp] Writing notebook server cookie secret to /root/.local/share/jupyter/runtime/notebook_cookie_secret
[W 19:26:58.951 LabApp] All authentication is disabled.  Anyone who can connect to this server will be able to run code.
[I 19:26:58.964 LabApp] JupyterLab extension loaded from /conda/envs/rapids/lib/python3.6/site-packages/jupyterlab
[I 19:26:58.964 LabApp] JupyterLab application directory is /conda/envs/rapids/share/jupyter/lab
[W 19:26:58.966 LabApp] JupyterLab server extension not enabled, manually loading...
[I 19:26:58.968 LabApp] JupyterLab extension loaded from /conda/envs/rapids/lib/python3.6/site-packages/jupyterlab
[I 19:26:58.968 LabApp] JupyterLab application directory is /conda/envs/rapids/share/jupyter/lab
[I 19:26:58.969 LabApp] Serving notebooks from local directory: /rapids/notebooks
[I 19:26:58.969 LabApp] The Jupyter Notebook is running at:
[I 19:26:58.969 LabApp] http://(712e75ae4a0e or 127.0.0.1):8888/
[I 19:26:58.969 LabApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[I 19:27:29.919 LabApp] 302 GET / (172.25.10.173) 1.71ms
[W 19:27:30.730 LabApp] Could not determine jupyterlab build status without nodejs
[W 19:27:30.925 LabApp] 404 GET /lab/api/workspaces/lab?1549654049120 (172.25.10.173): Workspace 'lab' ('lab-a511') not found
[W 19:27:30.925 LabApp] Workspace 'lab' ('lab-a511') not found
[W 19:27:30.926 LabApp] 404 GET /lab/api/workspaces/lab?1549654049120 (172.25.10.173) 1.45ms referer=http://172.25.10.206:8888/lab?

Screen shot - <host IP>:8888

To exit, select Shutdown from the File Menu:

Tensorflow:

NOTE: This will start Tensorflow container and switch to interactive console:

Command:

docker run --runtime=nvidia -it nvcr.io/nvidia/tensorflow:19.12-tf2-py3 bash

Please read the README.MD inside of the container for detail, or visit www.tensorflow.org for more information

Note:

Docker version earlier then 19.03 with nvidia-docker2 installed will need to use --runtime=nvidia flag for the NVIDIA GPU support in the container.

Docker version 19.03 and later with nvidia-container-toolkit installed will need to use ---gpus all flag for the NVIDIA GPU support in the container.

For Additional Technical Support, Please contact us at: www.exxactcorp.com/support