Decision note
Before scaling infrastructure, understand what actually needs to scale.
Scale problems are usually uneven. Compute, reads, writes, storage, queues and realtime telemetry do not grow at the same rate.
The operating view
Reduce the decision before increasing the technology.
Measure the pressure
Identify the resource that approaches its limit first and the business event that drives it.
Isolate high-frequency workloads
Telemetry, analytics and reporting should not automatically share the same scaling path as transactional operations.
Protect the critical path
Spend complexity on the workload whose failure stops the business, not on theoretical peak capacity everywhere.
Re-test economics
Architecture that scales technically but multiplies operational cost can still be the wrong design.