search

znmeb / edgyR

R on the Edge: NVIDIAⓇ Jetson™ tools for R developers
https://znmeb.github.io/edgyR/
GNU Affero General Public License v3.0
6 stars 4 forks source link

Separate build process into a builder image and a user image #3

Closed znmeb closed 4 years ago

znmeb commented 4 years ago

The current build process is monolithic - it starts with the NVIDIA base image and layers R and RStudio Server over the base. The proposal is to create a builder image with pure Ubuntu 18.04LTS as the base, containing R and RStudio Server built from source. This has a couple of advantages over the current process:

  1. R and RStudio Server change rarely - it's inefficient to re-compile them every time you want to release a new image.
  2. In theory you can use any NVIDIA L4T image as the "base" image for a release without recompiling R and RStudio Server.

Also, this very clearly delineates the open source base components - Ubuntu, R and RStudio Server - from the mixed-licensed components on an L4T image.

znmeb commented 4 years ago

This is working - refactoring at the moment. It looks like we'll have three images:

  1. A base image with Bionic (no NVIDIA), R and RStudio Server from source
  2. An image built on the NVIDIA base with R and RStudio binaries copied from the Bionic image, 3, An image built on the NVIDIA machine learning image - PyTorch, sci-kit learn, TensorFlow 1.15, etc.
znmeb commented 4 years ago

This is working - refactoring at the moment. It looks like we'll have three images:

  1. A base image with Bionic (no NVIDIA), R and RStudio Server from source
  2. An image built on the NVIDIA base with R and RStudio binaries copied from the Bionic image, 3, An image built on the NVIDIA machine learning image - PyTorch, sci-kit learn, TensorFlow 1.15, etc.
znmeb commented 4 years ago

Refactoring is complete - moving on to documentation

znmeb commented 4 years ago

Closing