7.4 Data Marts & Enterprise Data Warehouses

Data Marts, Enterprise Data Warehouses, and Integration Patterns

In enterprise data management, the configuration and integration of analytical data stores determine how effectively business units can access consistent data. An organization's architecture might feature a centralized Enterprise Data Warehouse (EDW), distributed data marts, a unified data hub, or a virtualized logical data layer. Understanding the distinctions, relationships, and governance challenges associated with these different configurations is essential for maintaining data integrity and avoiding toxic data silos.

Data Marts: Dependent vs. Independent

A data mart is a subset of data that is focused on a single subject area, department, or business function. Data marts are designed to make it easier and faster for specific user groups to access the data they need. However, the architectural design of a data mart dictates its long-term value to the enterprise.

Dependent Data Marts

Dependent data marts are populated directly from a centralized, unified Enterprise Data Warehouse (EDW). In this model (which aligns with Bill Inmon’s CIF architecture), all enterprise data is extracted from sources, cleaned, integrated, and loaded into the core 3NF EDW. The dependent data marts are then constructed by pulling data from this single source of truth and restructuring it to meet departmental needs.

• Pros: Complete data consistency. Because every data mart draws from the same source, their metrics, business definitions, and values are fully aligned. Reconciliation is straightforward, and enterprise-wide data governance is tightly controlled.
• Cons: Long initial implementation cycles. The centralized warehouse must be built first, meaning business departments may wait months before their specific data mart is operational. High upfront development cost.

Independent Data Marts

Independent data marts are built directly from operational source systems, bypassing any centralized data warehouse. Each department (such as Sales, Finance, or Marketing) designs and implements its own ETL pipelines, storage structures, and reporting systems to meet its immediate analytical needs.

• Pros: Rapid deployment and low initial cost for a single department. Business units have complete autonomy and do not have to wait for corporate IT to build an enterprise-wide system.
• Cons: Creation of highly fragmented data silos. Because there is no central coordination, departments define key metrics differently (for example, Sales and Finance may have different definitions of 'revenue' or 'active customer'). This leads to duplicate ETL pipelines, redundant storage, high maintenance costs, and conflicting reports that make corporate-level decision-making difficult. DAMA DMBoK2 strongly cautions against independent data marts due to these structural governance risks.

Data Hub Model

A data hub (or Data Hub Architecture) is a centralized model designed to facilitate data exchange, routing, and synchronization between multiple source and target systems. While it shares some physical similarities with a data warehouse (like centralized data storage), its purpose is entirely different.

• Data Hub: Focused on operational integration and near-real-time data exchange. It is a transit-oriented store that routes data between applications, services, and operational databases. It typically stores only the current, active state of the data, and data is often transient (purged after delivery).
• Data Warehouse (EDW): Focused on strategic analysis and historical reporting. It is a persistent store that keeps decades of historical data, structured for read-heavy query performance. It is not designed to route data back to operational systems in real-time.

Organizations often implement both. A data hub acts as the operational clearinghouse for app-to-app integration and master data synchronization, while periodically loading data into the EDW for long-term analytical tracking.

Virtual Data Warehouses

A Virtual Data Warehouse (VDW) is an architectural pattern that integrates data from disparate source systems using data virtualization technology instead of physically consolidating the data into a central database. The VDW establishes a logical middleware layer that exposes a unified semantic model. When a user runs a query, the virtualization engine translates it on-the-fly into multiple sub-queries, executes them against the physical source systems, compiles the results, and returns them to the user.

• Pros: Extreme agility. It allows organizations to prototype reports and access new data sources quickly without waiting for lengthy ETL developments. It avoids the hardware and storage costs of physically duplicating data.
• Cons: Performance is highly vulnerable to network latency and source database workloads; complex analytical queries can degrade operational systems. Furthermore, it does not preserve historical data; if a source system deletes records after 90 days, that historical data is permanently lost to the virtual warehouse.

Summary of Analytical Data Stores

CDMP Exam Tips and Common Pitfalls

• Disadvantages of Independent Data Marts: The CDMP exam will frequently test your knowledge of the risks of independent data marts. Look for terms like 'data silos,' 'redundant ETL,' 'reconciliation challenges,' and 'conflicting metrics.'
• Data Hub vs. EDW: Understand that a Data Hub is for operational exchange and is transactional/current-state, while an EDW is for analytical reporting and is historical.
• Data Virtualization Limits: Remember that data virtualization does not store physical data, making it unsuitable for long-term historical tracking when source systems delete historical records.