7 January 2026 · Wei-Han Lin
Latency budgets when half the estate stays on-prem
Hybrid cutovers fail quietly when chatty services cross the boundary without a budget. Here is how we set and defend those numbers.
Moving a customer-facing API while leaving its reporting database on-prem can look tidy on a slide. Under load, round-trips accumulate and operators spend the weekend chasing timeouts that were never budgeted.
We ask teams to name a latency budget per critical path before placement decisions harden. The budget forces a conversation about caching, batching, or keeping chatty pairs together until a later wave.
Budgets also protect rollback honesty: if measured latency exceeds the number you published, the cutover should stop — not negotiate mid-window.