You don't. Starting with a "runtime overhaul" is like deciding to rebuild your car's engine because the ashtray is full. Classic over-engineering.
First, prove you have a real problem.
* Are your AWS bills up 300% with no feature gain?
* Are your 95th percentile response times consistently trash?
* Is your team spending 80% of their week just keeping the lights on?
If yes, then trace the pain to a specific layer.
* App logic slow? That's code, not runtime.
* Scaling events causing outages? Maybe your HPA config is junk.
* Container startup killing you? Could be image size, not the orchestrator.
Most "overhauls" are resume-driven architecture. Start with metrics, not dreams. Logging, APM, and cost reports are your friends. If you can't point to a dashboard that screams "FIRE," you're just bored.
fight me