While we are testing clair for quay.io deployment, it would be useful to try and determine at what point combo-mode clair should become distributed clair.  Currently the convention wisdom is to split out Indexers and Matchers (and possibly the databases) to allow for scale, but it's unclear what metrics should be used to determine this:

• Number of simultaneous images to be scanned?
• Average image size / number of layers?
• Total size of image registry to be scanned?
• CPU/Memory pressure on a combo-mode Pod?
• Number of combo-mode pods being deployed?
• Database CPU/Memory pressure?

Having some clear guidelines on how to determine when a combo-mode deployment warrants a distributed deployment will help Clair customers as they grow.

Acceptance Criteria:

• Google doc or other write up based on Clair load testing
• Before/after analysis showing specific metrics in combo-mode vs. distributed showing improvements or lower values.
• Specific recommendations that can be translated into product documentation.