Skip to content

Creating Private Cloud Authorization Using On-premises Machines

You can run MissingLink jobs using on-premises machines. This topic describes how to turn one or more local machines into a computing grid for running Resource Management.

Preparation

Before you enable authorization on each specific machine, ensure that you perform the following procedures on each machine:

  • Install Docker:

    • For all machines types, follow the instructions here.
    • For a GPU machine, in addition to installing Docker, install the following:
  • Install Python.

  • Install and authenticate MissingLink's CLI as detailed here.
  • Configure AWS/GCP. This is required if you want to have access to your cloud and/or be able to access Data Volumes.

Note

For your convenience, a gist is available that contains the relevant portions of the script that we use for setting up our AMIs. The script installs Docker and NVIDIA drivers and pulls several useful images. The script assumes a clean Ubuntu installation with a user ubuntu but can be adapted to any platform supported by nvidia-docker.

Authorization

To authorize, on each machine, run:

ml resource local-grid init

Alternatively, use the following command:

ml resource local-grid init  --ssh-key-path PATH

where --ssh-key-path allows you to specify which SSH key to use instead of your default SSH key for SSH access running instances (can be changed per resource group), performing Git clone (can be changed per job) and encryption functionality.

Note

When running in hybrid environments, this same key must be used in every cloud or local instance in order to allow consistent decryption.