Docker

ML Docker Image installed on the Interaction Station ML computers (Ubuntu 16.04):

Installing Docker CE:

• sudo apt-get install apt-transport-https ca-certificates curl software-properties-common
• curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
• sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu xenial stable"
• sudo apt-get update

• More info: https://unix.stackexchange.com/questions/363048/unable-to-locate-package-docker-ce-on-a-64bit-ubuntu

Change Docker root dir using systemd (Don't do this, set volume instead)

• systemctl status docker.service
• sudo nano /etc/default/docker
• Edit ExecStart line to look like this ExecStart =/usr/bin/dockerd -g /media/MachineLearning/docker -H fd://
• systemctl daemon-reload
• systemctl restart docker
• sudo docker info - verify the root dir has updated
• https://github.com/IronicBadger/til/blob/master/docker/change-docker-root.md

Docker - clean up all the volumes

• sudo docker system prune -a -f --volumes

Installing nvidia-docker:

• #nvidia-docker2 still not supported by nvidia-docker-composite
• docker volume ls -q -f driver=nvidia-docker | xargs -r -I{} -n1 docker ps -q -a -f volume={} | xargs -r docker rm -f
• sudo apt-get purge -y nvidia-docker
• curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | \
• sudo apt-key add -
• distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
• curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | \
• sudo tee /etc/apt/sources.list.d/nvidia-docker.list
• sudo apt-get update
• sudo apt-get install -y nvidia-docker
• sudo pkill -SIGHUP dockerd
• #Test nvidia-smi with the latest official CUDA image
• docker run --runtime=nvidia --rm nvidia/cuda:9.0-base nvidia-smi
• Link:
• https://github.com/NVIDIA/nvidia-docker

Installing nvidia-docker-compose:

• pip install nvidia-docker-compose
• link: https://hackernoon.com/docker-compose-gpu-tensorflow-%EF%B8%8F-a0e2011d36
• Permission Denied on curl and save for docker compose: https://github.com/docker/machine/issues/652

Using Docker with nvidia-docker-compose

• Public docker repository (When doing FROM in Dockerfile, we need to select one of those)
• https://hub.docker.com/

• Dir structure:
• docker-compose.yml
• deepo
• deepo/do_not_finish.sh
• deepo/Dockerfile
• deepo_data (folder that is visible by deepo image)

• docker-compose.yml:

version: '3'
services:
  #machine name
  deepo:
    #container name
    container_name: deepo
    #path to Dockerfile
    build: deepo
    command: sh do_not_finish.sh
    volumes:
      - ./deepo_data:/media/deepo_data
    tty: true

• Dockerfile:

FROM ufoym/deepo ADD do_not_finish.sh /

• Dockerfiles guide:
• https://rock-it.pl/how-to-write-excellent-dockerfiles/

• do_not_finish.sh:

1. !/bin/bash

sh -c 'while :; do sleep 100; done'

• We need that endless loop, because docker-compose closes the container when is deployed
• The endless loop allowed us to use it with a docker exec

Run it

• Steps 1 and 2: Within the folder where is the docker-compose.yml file
• sudo nvidia-docker-compose build
• sudo nvidia-docker-compose up

• Step 3: From another terminal:
• sudo nvidia-docker exec -it deepo bash

Troubleshooting problems

• Check nvidia-docker version (needs to be version 1)
• nvidia-docker version
• More info:
• https://github.com/eywalker/nvidia-docker-compose/issues/26

• Permission denied: u'./docker-compose.yml
• https://github.com/docker/docker-snap/issues/26

Deepo

It includes:

• cudnn
• theano
• tensorflow
• sonnet
• pytorch
• keras
• lasagne
• mxnet
• cntk
• chainer
• caffe
• caffe2
• torch

Installing Deepo:

• Prerequisite 1: Nvidia driver
• Prerequisite 2: nvidia-docker
• Deepo (for GPU)

Run Deepo image with Docker:

• sudo nvidia-docker run -it ufoym/deepo:gpu bash

Run Deepo image with Docker (with python 2.7):

• sudo nvidia-docker run -it ufoym/deepo:py27 bash

Setting up ML computers:

• Linux distribution installed: Ubuntu 16.04

Partition made for machine learning:MachineLearning

• In Windows: Disk Management -> Resize DataStorage
• Create new ext4 patition

Mounting the partition automatically:

Get the UUID of the learning:MachineLearning partition

• sudo blkid

Add partition to fstab:

• sudo nano /etc/fstab
• Add at the bottom these two lines:
• UUID=(id of the MachineLearning partition) /media/MachineLearning rw,suid,dev,auto,user,async,exec 0 2
• UUID=(id of the DataStorage partition) /media/DataStorage ntfs-3g defaults=en_US.UTF-8 0 0

• The parameters mounting the MachineLearning partition solved this problem running caffe from that partition:
• https://github.com/rbgirshick/py-faster-rcnn/issues/162
• https://askubuntu.com/questions/678857/fstab-doesnt-mount-with-exec

Give writing permissions to new MachineLearning partition

• sudo chmod -R a+rwx /media/MachineLearning/

• Need extra space? Extending the partition

https://askubuntu.com/questions/492054/how-to-extend-my-root-partition

Installing NVIDIA Driver:

• Set Ubuntu to boot on console mode. Type:
• sudo apt-get install systemd
• sudo systemctl set-default multi-user.target
• sudo reboot now
• Login and in console mode, type:
• sudo add-apt-repository ppa:graphics-drivers/ppa
• sudo apt update
• sudo apt upgrade
• For GeForce 1070Ti (07/2018), type:
• sudo apt-get install nvidia-390
• Re-set Ubuntu to boot on graphical mode. Type:
• sudo systemctl set-default graphical.target
• sudo reboot now

Checking if Nvidia Driver is properly installed. Type:

• nvidia-smi
• nvidia-settings

Installing CUDA 9.0 for Ubuntu 16.04 (the latest version is not supported by TensorFlow):

• wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
• wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64/libcudnn7_7.0.5.15-1+cuda9.0_amd64.deb
• wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64/libcudnn7-dev_7.0.5.15-1+cuda9.0_amd64.deb
• wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64/libnccl2_2.1.4-1+cuda9.0_amd64.deb
• wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64/libnccl-dev_2.1.4-1+cuda9.0_amd64.deb
• sudo dpkg -i cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
• sudo dpkg -i libcudnn7_7.0.5.15-1+cuda9.0_amd64.deb
• sudo dpkg -i libcudnn7-dev_7.0.5.15-1+cuda9.0_amd64.deb
• sudo dpkg -i libnccl2_2.1.4-1+cuda9.0_amd64.deb
• sudo dpkg -i libnccl-dev_2.1.4-1+cuda9.0_amd64.deb
• sudo apt-get update
• sudo apt-get install cuda=9.0.176-1
• sudo apt-get install libcudnn7-dev
• sudo apt-get install libnccl-dev
• sudo reboot now
• export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}
• export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
• sudo nano .bashrc
• Add the two last export lines at the end of the file. Save and reboot.

Checking if CUDA is properly installed. Type:

• nvcc --version

Resources used:

• https://askubuntu.com/questions/61396/how-do-i-install-the-nvidia-drivers
• https://medium.com/@bbloks/a-machine-learning-environment-with-ubuntu-and-gpu-acceleration-in-5-steps-765608325356
• https://yangcha.github.io/CUDA90/

Other options:

NTFS fstab wizard:

• sudo apt-get install ntfs-config
• sudo ntfs-config

Format large capacity HD with fs ExFat for having access to it from Ubuntu:

• On Windows 10
• cmd
• diskpart
• select disk '#' (where # is the number of the target drive)
• list part
• select part # (where # is the number of the partition)
• format fs=exfat QUICK