Introduction

Running OpenShift involves 3 kinds of costs:

• Cost of the user workloads
• Cost of the platform itself (e. g. control plane)
• Cost of the remaining idle capacity on worker nodes

Each user should be charged for its share of the OpenShift cluster, including all three of those costs.

Control plane

OpenShift requires a minimum of 3 master nodes for the control plane and 2 worker nodes.

Master nodes do not run user workloads so it’s an overhead when the cluster is dedicated. 

• This is not a problem (for users) on managed offerings such as ARO, ROSA, etc where Red Hat pays for the control plane
• Conversely, there is also the HyperShift case, where a control plane is shared across multiple clusters and each one of them should get its bill.

User workloads

Worker nodes run user workloads but they also run other processes which are perceived by users as overhead:

• Operating system
• Native OpenShift projects

Idle capacity

Another “overhead” is the idle capacity. There are 2 main causes for idle capacity:

• Mr Customer requested more of resource X (CPU, RAM, etc) than he needs. This is already accounted for by the current cost models in a variety of ways (requested, effective, etc) and it’s users who pay for this. Nothing to do here.
• The cluster is oversized (vs its used capacity), therefore there is some idle capacity that could be used by future user projects but at the moment is not being used. Someone must pay for this:
  • Central IT department?
  • Users proportionally to CPU or RAM or …?
  • Fixed percentage per user of that cluster?
  • Other strategy?

More

More details in these work documents:

https://docs.google.com/document/d/1-XL1vFGp1w0U2QIwT9OSkQz-EqKrodeqYngkUsoUh7o/edit#

https://docs.google.com/document/d/1Bm_4Ji3QRjZSuoHlauYdnQmv-E2bG9p8K-ICXcJ8QjQ/edit#heading=h.va450hxhd2kq