06-reference

dedp semantic layer bi olap virtualization

Fri Apr 03 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·book-chapter ·source: https://www.dedp.online/part-2/4-ce/semantic-layer-business-intelligence.html ·by DEDP / Simon Späti

DEDP 4.6 — BI, Semantic Layer, Modern OLAP, Data Virtualization

Four technologies converging on one goal: translate raw data into business meaning at query time. The chapter traces the evolution from embedded BI dashboards to standalone semantic layers, modern OLAP engines, and virtualization — all solving the abstraction problem differently.

The Four Systems

Business Intelligence Dashboard/Report: The oldest and most visible layer. Roll-up for executive KPIs (the “cockpit” metaphor), drill-down for detail, self-service for analysts. Traditional BI required pre-built star schemas and manual joins inside the tool. Modern BI shifts toward declarative, lightweight, metric-centric interfaces.

Semantic Layer: A logical translation layer between raw data and business users. Maps metrics to physical tables declaratively. The concept is not new — SAP BusinessObjects patented it in 1991 as “Universe.” What changed: semantic layers decoupled from BI tools and became standalone (MetriQL, MetricFlow, Minerva, Cube). Key timeline:

The chapter draws a strong parallel to MVC architecture and Active Record ORM — the semantic layer is the Model that decouples data (database) from presentation (dashboard). This is the same insight as 06-reference/2026-04-03-headless-bi: separate the metric definition from the visualization tool.

Modern OLAP (Druid, Pinot, ClickHouse, Kylin): Next-gen analytical engines that handle business logic at query time, unlike traditional OLAP cubes that pre-process everything. Key difference: measures and queries are defined dynamically, not baked into cube structures. These systems blur into semantic layers — they are essentially “OLAP cubes extended with access permissions, API layers, and data modeling features.” Support real-time streaming alongside batch.

Data Virtualization / Federation (Dremio, Trino, Denodo): Query multiple heterogeneous sources without moving data. Creates logical abstraction layers over source systems. Dremio’s “Data Reflections” patent creates database-like indexes on source systems for better performance than simple query pushdown. Trade-offs: slower than materialized queries, risk of impacting source system performance, data quality must be handled at source or query time.

Shared Patterns

All four systems share:

  1. Abstraction / Translation — convert technical structures to business terms
  2. Data Modeling — connect business concepts to physical structures
  3. SQL / YAML as interface — declarative metric definition
  4. Speed / Caching — fast retrieval through various caching strategies
  5. Single Source of Truth — consistency across consumers
  6. Metrics / KPIs as the core value proposition

Two design patterns emerge explicitly: ad-hoc querying (flexible, evolving questions without predefined constraints) and caching (persist for speed, accept staleness trade-off).

What Matters for Consulting

The semantic layer is the strategic bet. This chapter reinforces what 06-reference/2026-04-03-headless-bi argues: the semantic layer is where metric governance lives. It is the control point. Every 01-projects/phdata/index client building a “modern data stack” will eventually need a semantic layer, whether they call it that or not.

MVC parallel is a powerful teaching tool. Engineers understand MVC intuitively. Telling a client “your semantic layer is the Model in MVC — it decouples your data from your dashboards the same way Rails decouples your database from your views” lands immediately. Use this in discovery conversations.

Modern OLAP vs. semantic layer distinction matters for architecture decisions. The chapter says these are blurring — and that is true. But for clients: if your primary need is sub-second aggregation on streaming data, modern OLAP (ClickHouse, Pinot). If your primary need is consistent metric definitions across tools, semantic layer (Cube, MetricFlow). Different entry points, converging destinations.

Data virtualization is underused in enterprise. Most clients default to “move everything into the warehouse.” Virtualization is the right answer when: (a) source systems are authoritative and well-governed, (b) data freshness requirements are real-time, (c) the governance cost of copying exceeds the performance cost of federation. Relevant for 01-projects/data-marketplace/index — federated access to data products without centralized copying aligns with data mesh principles.

The 1991 origin story kills the “semantic layer is new” narrative. BusinessObjects Universe was a semantic layer 35 years ago. This is useful ammunition when clients or vendors present semantic layers as novel. The pattern is old; the standalone implementation is new. 06-reference/concepts/analytics-as-craft — knowing the history prevents reinventing it.

What’s Academic

The Data Virtualization section is the weakest — it lists tools without deeply analyzing when virtualization outperforms materialization. The “Dremio Reflections” detail is interesting but niche.

Key Takeaway

The semantic layer is the architectural control point for metrics governance. Whether it lives inside BI, inside OLAP, or standalone, the pattern is the same: define metrics once, serve them everywhere. The tools converge; the pattern endures. 06-reference/concepts/systems-over-goals — build the metric definition system, not the dashboard.