Chapter 2. Process Description

Like most Red Hat OpenStack Platform features, hyper-convergence is best implemented through the director. This allows you to take advantage of existing Heat templates and environment files to orchestrate your deployment. In particular, the director also includes a default environment file for implementing a supported version of hyper-convergence.

On the other hand, the director’s infrastructure also provides a framework you can use to define your own Heat templates and environment files. This is useful when the existing ones do not cover your specific use case.

The following subsections briefly describe each step of the deployment process.

Choosing a Deployment Type

Red Hat OpenStack Platform allows you to deploy two types of HCIs: pure and mixed. For both types, Red Hat Ceph Storage is deployed as the primary storage provider for Red Hat OpenStack Platform services.

On pure HCI, all Compute nodes will be deployed with Ceph-OSD services co-located on them.

On mixed HCI, you can deploy normal Compute nodes alongside those with co-located Ceph-OSD services.

Before you start your deployment, determine which type of HCI you are deploying, as their respective implementations are quite different.

Configuring Resource Isolation
When you deploy HCI, the director does not configure Compute and Ceph Storage services to be aware of each other on hyper-converged nodes. Both services will consume resources as if they were on dedicated nodes. This can lead to resource contention, which can lead to performance degradation. You need to mitigate this manually.
Configure Networking
For both Pure and Mixed HCI deployments, you need to map the StorageMgmtNetwork ports to the right NICs. During this step, you can implement any other networking settings required in your environment.
The deployment process for HCI (whether pure or mixed) involves specifying which environment files to include in the deployment. Pure HCI involves specifying an existing environment file that configures the services included in the Compute role. Mixed HCI involves defining a new flavor, tagging it to hyper-converged nodes, and invoking a custom roles file during deployment.