Column-Oriented Storage

Analytic queries often compute aggregates over large datasets. Reading entire rows when you only need one column is wasteful.

Column Compression

Storing all values of a column together allows for extreme compression:

• Bitmap Encoding: For low cardinality columns (e.g., "Gender"), store a bit for each value.
• Run-Length Encoding: Compress repetitive sequences.

Vectorized Processing

Column stores can use CPU SIMD (Single Instruction, Multiple Data) instructions to process chunks of column data in tight loops without function calls.

Materialized Aggregates

If the same aggregates (SUM, COUNT) are used frequently, we can cache them:

• Materialized Views: A table-like object containing the actual results of a query, automatically updated when the underlying data changes.
• Data Cubes (OLAP Cubes): A grid of precomputed aggregates grouped by different dimensions (e.g., product, date, region). Queries become extremely fast because results are pre-calculated.