Data Warehouse
Subject: Computer and Technology noun /ˈdeɪ.tə ˈwɛrˌhaʊs/ 1 definition
- data-warehousing
- database-systems
- data-engineering
- analytical-processing
-
Definition (noun)
Entry #307 · UpdatedA Data Warehouse is a centralized, subject-oriented repository designed to consolidate, store, and analyze large volumes of historical data from multiple heterogeneous operational sources. Records enter the warehouse through scheduled batch or streaming ETL (Extract, Transform, Load) and ELT pipelines that cleanse, deduplicate, and conform disparate datasets into structured analytical schemas—most commonly star or snowflake models built around fact and dimension tables. To sustain high throughput for online analytical processing (OLAP) queries, modern data warehouses use columnar storage layouts, partition pruning, and massively parallel processing (MPP) engines to aggregate billions of rows efficiently. Unlike operational OLTP databases engineered for high-concurrency, low-latency row mutations—or data lakes that store raw, uncurated files—a data warehouse enforces governed relational schemas and non-volatile historical retention tailored for SQL analytics and business intelligence.
Common Types
Enterprise Data Warehouse (EDW): A centralized, organization-wide repository that unifies governed historical data from across all business units into a single authoritative schema for cross-functional reporting and compliance.
Cloud Data Warehouse: A managed platform built on cloud object storage and decoupled compute clusters that scales storage capacity and query concurrency independently on demand (such as Snowflake, Google BigQuery, and Amazon Redshift).
Data Mart: A subject-oriented warehouse subset scoped to an individual department such as finance or sales—either derived from a central EDW or loaded directly from source systems—to simplify schemas and speed up domain queries.
Virtual Data Warehouse: A logical federation layer of views and distributed query connectors that executes analytical SQL across underlying operational databases in place without copying records into separate physical storage.
Example
Every night at 02:00 UTC, an Apache Airflow pipeline extracts daily transactions from PostgreSQL and ad-spend metrics via a REST API, transforms them with dbt into a
fact_ordersstar-schema table in Snowflake, and appends 14.2 million conformed rows. Analysts then execute multi-yearSELECT region, SUM(revenue) FROM fact_orders GROUP BY regionqueries in under three seconds without impacting the latency of the live checkout database.