Row vs. Column: How Storage Layout Shapes Everything

This article is Part 2 of a 10-part series on query engine design, focusing on the fundamental decision of how bytes are arranged on disk. It details how row storage (used by PostgreSQL, MySQL, Oracle) keeps all fields of a record together, optimized for transactional workloads but wasteful for analytical queries. Column storage (used by DuckDB, ClickHouse, Snowflake) keeps values of each field together, enabling efficient I/O and superior compression for analytics. It covers I/O math, late materialization, hybrid approaches, and real-world system choices, providing guidance on when to choose each layout.