General Process for Docker Container

General-purpose processing for (1) docker image pulling, (2) docker building, (3) docker container running.

1. Pull a docker image

sh docker_pull.sh

Specify the image name to the proper CUDA and Ubuntu version you want at dockerhub_nvidia.

• IMG_NVDA: Name of the docker image by NVIDIA

2. Build the image

sh docker_build.sh

Specify your image name and build the image using the nvidia image via Dockerfile.

• IMG_NVDA: Name of the docker image by NVIDIA which you pulled above
• IMG_NAME: Name of your image to be built

3. Docker run

sh docker_run.sh
sh docker_run_x11.sh # if you use x11 forwarding

Specify your preference with the environment variables:

• CONT_NAME: Name of the container to be created
• VOL_DATA: Linking any path to dataset for training (maybe different to each server)
• VOL_RVI, VOL_RVI2: Linking the lab dataset folder of the storage server to the specificed container path
• VOL_MINE, VOL_MINE2: Linking your dataset folder of the storage server to the specificed container path
• VOL_DEV: Linking your dev folder of the node to the specificed container path
• PORT_TB: Port number of the Tensorboard (6006)
• PORT_SSH: Port number of the SSH (22)
• PORT_JUP: The first port number of the jupyter (8888)
• PORT_JUP_: The second port number of the jupyter (8889)
• IMG_NAME: Name of the image you built above