This section is of concern if you are Deploying Substra in production.
Walkthrough guides you through deploying a production environment
How-to are guides on some tasks of interest.
Upgrade notes should be read when upgrading from one version to the next.
Compatibility table holds a reference of Substra versions.
How Substra is deployed¶
Substra is meant to be deployed as part of a federated learning network. Each participant organization will set up their own Substra node, from which their users can connect to the network and run machine learning algorithms on the data registered by participant on their own node.
The terms Substra node and Substra organization are practically interchangeable.
Substra is distributed as Helm charts, running on Kubernetes 1.19 and up. Each component has their Helm chart, which are hosted at https://substra.github.io/charts.
Each backend needs the following resources to run Substra:
30 GB of RAM
300 GB of storage
In addition, you need to consider the resources required by the compute tasks. For example, if each task needs 10 GB of RAM and you have two tasks running in parallel for a single backend, you will need a total of 50 GB of RAM (30 GB + 2*10 GB). The same applies to CPU usage and storage requirements (datasets and models).
The orchestrator needs the following resources:
16 GB of RAM
100 GB of storage