Hub-and-spoke is Bill Inmon's blueprint for connecting a warehouse's layers. A single normalized enterprise data warehouse sits at the center - the hub - holding integrated, historical, atomic data in third normal form. Dimensional data marts radiate from it as spokes, each one a star-schema projection of the hub built for a specific subject area or department. Source data flows into the hub once, is integrated and conformed there, and is then delivered out to the spokes. The name is literal: one center, many arms.

How data moves through it

The flow is one-directional and staged. Operational systems load into the normalized hub, where the warehouse resolves duplicate keys, reconciles conflicting definitions, and stores the single integrated version of each entity. From the hub, ETL jobs populate each dimensional mart with the slice it needs, denormalized into facts and dimensions for fast querying.

source systems  →  normalized EDW (hub, 3NF)  →  dimensional marts (spokes, star schemas)
                         │                              ├─ sales mart
                         │                              ├─ finance mart
                         └─ single integrated copy      └─ supply-chain mart

Because every spoke descends from the same hub, the marts are dependent marts - "customer" and "revenue" carry identical meaning across all of them, since all of them inherit those definitions from the one place the integration happened.

Versus the bus architecture

Kimball's bus architecture reaches the same goal - consistent enterprise reporting - without the normalized hub. Instead of integrating into a central 3NF store first, it builds dimensional marts directly and ties them together with conformed dimensions: shared dimension tables (a single dim_date, dim_customer, dim_product) used identically by every fact across every mart. The conformance lives in the shared dimensions rather than in a separate normalized layer.

	Hub-and-spoke (Inmon)	Bus (Kimball)
Center of gravity	Normalized EDW hub	Conformed dimensions
Marts are	Dependent on the hub	Built directly, sharing dimensions
Integration happens	In the hub, before marts	In the conformed dimensions
Extra storage layer	Yes (the 3NF hub)	No separate normalized store

For the methods these architectures belong to, see the Inmon method and the Kimball method, and the broader EDW versus data marts trade-off.

Frequently asked questions

What is hub-and-spoke architecture? Inmon's design where a normalized enterprise warehouse (the hub) feeds dimensional data marts (the spokes), so integration happens once in the center before data is delivered out.

What's the hub? The central enterprise data warehouse, normalized to third normal form, holding the single integrated, historical copy of all subject areas.

How is it different from the bus architecture? Hub-and-spoke integrates into a normalized central store first; the Kimball bus integrates through conformed dimensions shared across marts, with no separate normalized layer.

Are hub-and-spoke marts dependent or independent? Dependent - they're sourced from the hub, so all marts share the hub's definitions and keys.