7.1 DW Architectures: Inmon vs Kimball

Overview of Data Warehousing Architectures

Within the discipline of Data Warehousing (DW) and Business Intelligence (BI), the architecture of the analytical data environment is the foundational factor that determines scalability, governance, development speed, and data consistency. A Data Warehouse is defined by the Data Management Body of Knowledge (DAMA DMBoK2) as a database structured to support query and analysis. It acts as an integrated repository of historical data collected from multiple operational systems. The historical debate surrounding data warehouse design centers on two primary schools of thought: the top-down methodology pioneered by Bill Inmon, and the bottom-up methodology championed by Ralph Kimball. Both methodologies aim to deliver a single version of the truth, but they differ fundamentally in data structuring, ingestion pipelines, integration mechanisms, and organizational deployment models.

Bill Inmon's Corporate Information Factory (CIF)

Bill Inmon’s architecture is structured around the concept of the Corporate Information Factory (CIF). In this model, data warehousing is approached from an enterprise-wide, top-down perspective. The central component is the Enterprise Data Warehouse (EDW), which stores atomic-level data in Third Normal Form (3NF). This database is highly normalized to minimize data redundancy, protect data integrity, and enforce strict, centralized business rules.

In the CIF architecture, the ingestion flow starts by extracting data from operational source systems, cleansing it in a staging area, and loading it into the normalized 3NF EDW. Because the EDW is normalized, it is not optimized for direct user querying or analytical reports due to the complex multi-table joins required. To facilitate reporting, specialized dependent data marts are created downstream. These data marts represent physical subsets of the EDW that are restructured into dimensional formats (like star schemas) tailored for specific business departments (e.g., Finance, HR, Sales). Because these data marts are derived directly from the centralized EDW, data consistency is maintained across the entire enterprise.

• Key Philosophy: The warehouse is a centralized, normalized repository of enterprise data. Data marts are downstream, dependent representations.
• Data Model: Entity-Relationship modeling in Third Normal Form (3NF) for the core repository, and dimensional modeling for the derived data marts.
• Pros: Extreme consistency, low data redundancy, centralized governance, and high flexibility for structural changes in source systems.
• Cons: High upfront costs, slow time-to-market, requires complex ETL workflows to populate the 3NF repository and then rebuild dependent data marts.

Ralph Kimball's Dimensional Data Warehouse (DDW)

In contrast, Ralph Kimball’s architecture is built on the Dimensional Data Warehouse (DDW), also known as the Data Warehouse Bus Architecture. Kimball views the data warehouse not as a single physical repository, but as the collective union of all dimensional data marts across the enterprise. It uses a bottom-up, process-centric approach where data is modeled dimensionally from the very beginning.

Instead of a central 3NF database, Kimball’s architecture relies on the Data Warehouse Bus. This bus is a conceptual framework that defines shared, standardized dimensions called conformed dimensions (such as Customer, Product, Date, and Location) and standardized fact tables. A conformed dimension is a dimension table that has the exact same structure, keys, and values across different data marts. By using conformed dimensions, different business process data marts can be integrated seamlessly. Users can perform cross-functional queries (known as 'drilling across') because the dimensional attributes are identical across the sales, inventory, and billing datasets.

• Key Philosophy: The data warehouse is the union of all dimensional data marts, integrated via conformed dimensions on a shared bus.
• Data Model: Dimensional modeling using star schemas (facts and dimensions) at the atomic level.
• Pros: Faster implementation, lower initial costs, highly intuitive for business query tools, and optimized for fast read performance.
• Cons: Requires rigorous enterprise-wide data governance to maintain conformed dimensions, higher data redundancy, and can be difficult to restructure if business processes change.

Comparison of Inmon and Kimball Architectures

Hub-and-Spoke vs. Bus Architecture

The choice between Inmon and Kimball is often framed as a choice between Hub-and-Spoke Architecture and Bus Architecture:

1. Hub-and-Spoke Architecture: This represents the physical implementation of Inmon's CIF. The 'Hub' is the centralized 3NF EDW. The 'Spokes' are the dependent data marts. All data flows through the hub before reaching the spokes. This model ensures that no department can create a localized 'truth' that contradicts the enterprise standard. It provides an excellent control point for data governance, metadata management, and security, but can suffer from throughput bottlenecks.
2. Bus Architecture: This represents Kimball's DDW. There is no physical central hub database. Instead, the bus is a virtual backplane where dimensional data marts are physically separate but logically connected through conformed dimensions. The architecture is highly distributed and scalable, allowing different business units to build their marts independently, provided they conform to the enterprise bus matrix. The challenge is that without active governance, departments may drift and modify conformed dimensions, creating data silos.

CDMP Exam Tips and Common Pitfalls

When preparing for the CDMP exam, keep the following distinctions in mind:

• Normalized vs. Dimensional: If an exam question describes a data warehouse as a 'normalized relational database serving as the single source of truth,' it is referring to Inmon's CIF. If it refers to 'conformed dimensions, facts, and star schemas,' it is Kimball's DDW.
• Dependent Data Marts: Remember that dependent data marts are physical databases downstream of a centralized EDW. They are associated with Inmon. In Kimball's architecture, data marts are the building blocks of the data warehouse itself and are connected via the bus.
• Bus Matrix: A key tool in Kimball's methodology is the Data Warehouse Bus Matrix. This matrix lists the business processes as rows and the conformed dimensions as columns, indicating which dimensions are used by which processes. This is a common exam concept for planning dimensional warehouses.