Data Virtualization and Logical Data Management Fundamentals
What this domain tests
This domain checks whether you can reason about data virtualization as an architecture: a logical model over heterogeneous sources, exposed as views, queried at run time. The exam expects precise use of terms such as base view, derived view, interface view, query delegation, and logical data warehouse. You should be able to choose virtualization versus replication and explain the trade-offs without assuming that virtualization means “no storage anywhere.”
Core rules
- A view is a named, typed query, not a physical table. A base view wraps a source object; a derived view combines or transforms other views; an interface view publishes a stable contract for consumers.
- Logical data management separates the consumer-facing schema from source schemas. Consumers query entities and relationships; the platform maps them to sources and reconciles dialects.
- Data remains in systems of record by default. Filters, joins, and aggregations may be delegated to sources; the virtualization server evaluates what cannot be pushed down and combines results.
- Abstraction and encapsulation are the point: source changes should be absorbed by remapping a view, not by rewriting every report.
- Virtualization and ETL are complementary. ETL copies and transforms at rest; virtualization federates and transforms on read. Caching or materialization is a deliberate performance or availability decision.
Worked example
A retailer needs a Customer 360 view from CRM (customer profile), ERP (orders), and a loyalty SaaS (points). Create three base views by introspection. Create a join view on customer_id, then an interface view exposing customer_id, name, segment, lifetime_value, and points_balance. A BI query filtering segment = 'Gold' is delegated to CRM; the order aggregation is pushed to ERP if supported; loyalty data is joined in the platform. No nightly copy is created, and each source remains the system of record. If loyalty latency is unacceptable, add a cache with a defined refresh policy rather than silently copying data into a new warehouse.
Traps that cost marks
- Saying virtualization never stores data: cache, temporary results, and MPP execution can store data.
- Treating a view as materialized by default.
- Assuming every SQL construct is pushed down; unsupported functions are evaluated in the platform.
- Believing virtualization removes the need for security and lineage at the view layer.