Better layering of docker image

[ThomasDelteil]

Currently the Dockerfile use a single layer to do an awful lot of things, which means any changes on this line

RUN apt-get update && \
    apt-get install -y software-properties-common && \
    add-apt-repository -y ppa:certbot/certbot && \
    apt-get update && \
    apt-get install -y \
    build-essential \
    libatlas-base-dev \
    libopencv-dev \
    graphviz \
    python-dev \
    openjdk-8-jdk \
    nginx \
    protobuf-compiler \
    libprotoc-dev \
    python-certbot-nginx \
    curl \
    vim \
    && rm -rf /var/lib/apt/lists/* \
    && cd /tmp \
    && curl -O https://bootstrap.pypa.io/get-pip.py \
    && python get-pip.py \
    && pip install gevent gunicorn mxnet-model-server \
    && pip uninstall --yes mxnet \
    && pip install mxnet-mkl \
    && mkdir /mxnet_model_server

triggers the creation of a new heavy layer. I would suggest separating the dependencies that are not likely to change and make it a base image.
Then you can have a different image for cpu/gpu that mostly only installs the right version of mxnet and mxnet-model-server. This makes the updates to the awsdeeplearningteam/mms_cpu much faster. I am writing this while waiting for the updates to the latest version that I hope might fix the bug I reported earlier:

Step 1/4 : FROM awsdeeplearningteam/mms_cpu:latest
latest: Pulling from awsdeeplearningteam/mms_cpu
22dc81ace0ea: Already exists
1a8b3c87dba3: Already exists
91390a1c435a: Already exists
07844b14977e: Already exists
b78396653dae: Already exists
8f4e3302427d: Downloading [========================================>          ]  515.4MB/639.9MB
8f4e3302427d: Downloading [=========================================>         ]  532.1MB/639.9MB

Actually I am going to issue a PR

edit: holding off on the PR, since I have seen the Dockerfile is different here and pin-pointing the mxnet-model-server version: https://hub.docker.com/r/awsdeeplearningteam/mms_cpu/~/dockerfile/

[vdantu]

Reg the layering, thats a good point. We will break this up further.

[vrakesh]

Added in pull request #363