Top-down enterprise data warehouse architecture
The Inmon Methodology, developed by Bill Inmon (often called the "Father of Data Warehousing"), is a top-down approach that emphasizes building an integrated, centralized Enterprise Data Warehouse (EDW) as the single source of truth for the entire organization.
This approach focuses on creating a normalized, subject-oriented data warehouse at the enterprise level first, from which departmental data marts are derived. The methodology prioritizes data integrity, consistency, and long-term architectural stability over rapid delivery.
Build the Enterprise Data Warehouse first as a comprehensive, integrated repository. Once the EDW is established, create dependent data marts by extracting subsets for specific departments or business functions.
Organize data around major business subjects (Customer, Product, Transaction) rather than applications or departments. Each subject area is comprehensively modeled with all relevant attributes and relationships.
The EDW uses normalized data models (typically 3NF - Third Normal Form) to eliminate redundancy, ensure data integrity, and provide flexibility for future requirements. Denormalization occurs in data marts for query performance.
The EDW serves as the authoritative, integrated repository for all enterprise data. All reports, analytics, and data marts derive from this central source, ensuring consistency across the organization.
The Inmon architecture consists of distinct layers, each serving a specific purpose in the data flow from source systems to end users:
Operational systems, legacy applications, external data sources, and third-party systems. Data originates here in various formats and structures.
Temporary storage for data extraction and initial transformation. Includes data quality checks, cleansing, and preparation before loading into the EDW.
The core integrated repository with normalized data models. Contains comprehensive, historical, integrated data organized by subject areas. This is the single source of truth.
Departmental or functional subsets derived from the EDW. Denormalized for query performance and tailored to specific user groups (Sales, Finance, Marketing marts).
| Aspect | Inmon (Top-Down) | Kimball (Bottom-Up) |
|---|---|---|
| Approach | Enterprise warehouse first, then data marts | Data marts first, integrated via conformed dimensions |
| Data Model | Normalized (3NF) in EDW | Denormalized (Star/Snowflake) |
| Focus | Subject-oriented, enterprise integration | Business process-oriented, rapid delivery |
| Time to Value | Longer initial implementation (months to years) | Faster (weeks to months per data mart) |
| Complexity | Higher initial complexity, comprehensive scope | Lower per-iteration complexity |
| Flexibility | High for unforeseen requirements | Moderate, optimized for known queries |
| Query Performance | Slower in EDW, fast in data marts | Fast for typical analytical queries |
| Data Integrity | Strong enforcement through normalization | Managed through ETL processes |
| Best For | Large enterprises with complex integration needs | Organizations needing quick wins and iterative delivery |
The Inmon approach is particularly well-suited for: