BDC Sizing & Capacity

As of 2026-05-29

Sizing BDC = enough capacity for the workload without paying for unused headroom; in a consumption model (M032) wrong-sizing is expensive both ways (under = slow/SLA-miss/distrust, over = permanent bill fat). Workload-driven, not seat-driven: three classes scale on different axes — scheduled pipelines (batchable), interactive queries/stories (concurrency/latency), Databricks eng/ML (spiky, M034); one number for all three is the classic error. T-shirt estimate first, then right-size on real telemetry (M045) after 4-6 weeks — don't treat the initial guess as final. Design for elasticity (scale up for month-end close, down after) not standing peak capacity. Storage sizing follows the history tiering (M071) — size hot+warm working set, not the 12-year archive. Honest (§0i): sizing is an estimate refined by reality + a right-sizing checkpoint, not false-precision. Controls bill + UX simultaneously.