Michael J. Swart does some modeling:
How do we predict whether it’s safe to put workloads from two servers onto one?
We use Availability Groups to create readable secondary replicas (which I’ll call mirrors). The mirrors are used to offload reporting workloads. The mirrors are mostly bound by IOPS and the primaries are mostly bound by CPU, so I wondered “Is there any wiggle room that lets us consolidate these servers?”
Can we point the reporting workloads (queries) at the primary replica safely?
Read on for the answers to these questions.