How do columnar databases optimize for analytical workloads?

Columnar databases (Redshift, BigQuery, ClickHouse) store data by column instead of row. Benefits: read only needed columns (reduced I/O), better compression (similar values together - RLE, dictionary, delta), vectorized processing (SIMD on column batches), zone maps/min-max indexes (skip blocks). Trade-offs: row reconstruction for wide reads, slower single-row operations, complex updates. Perfect for aggregations, scans on few columns. Row stores better for OLTP with full-row access.