This reference architecture describes how to deploy Red Hat® OpenStack® Platform and Red Hat Ceph Storage in a way that both the OpenStack Nova Compute services and the Ceph Object Storage Daemon (OSD) services reside on the same node. A server which runs both compute and storage processes is known as a hyper-converged node. There is increasing interest in the field for hyper-convergence for cloud (NFVi and Enterprise) deployments. The reasons include smaller initial deployment foot prints, a lower cost of entry, and maximized capacity utilization.

The first section of this reference architecture provides a technical summary for an implementer to quickly deploy a hyper-converged overcloud by using templates and scripts from this reference architecture on GitHub.

The second section provides general hardware and network guidance.

The third section covers software prerequisites for the undercloud and managing hardware with Ironic.

The fourth section covers how to define a hyper-converged overcloud in Heat which is stored in a deployment plan.

The fith section covers how to isolate resources in a hyper-converged overcloud to address contention between OpenStack and Ceph which could result in degradation of either service.

The sixth section covers how to deploy a hyper-converged overcloud using the deployment plan defined in the previous sections.

The seventh section covers some of the operational concerns of running a hyper-converged OpenStack and Ceph deployment. It covers configuration updates, adding new Compute/Red Hat Ceph Storage nodes, and removing running Compute/Red Hat Ceph Storage nodes.

This reference architecture has been completed with Red Hat Enterprise Linux® 7.3, Red Hat OpenStack Platform 10, Red Hat OpenStack Platform director 10, and Red Hat Ceph Storage 2. All of the steps listed were performed by the Red Hat Systems Engineering team. The complete use case was deployed in the Systems Engineering lab on bare metal servers, except where otherwise noted.