Star Schema Benchmark (SSB)

The Star Schema Benchmark (SSB) is a data-warehouse benchmark derived from TPC-H but restructured into a textbook star schema. It was introduced in academic work to study performance on the dimensional models that real warehouses actually use, where a central fact table joins to several denormalized dimension tables. It is widely used to evaluate columnar engines and OLAP databases because its shape matches common warehouse design.

What It Measures

SSB measures query execution time for 13 queries organized into four "flights," each flight progressively adding dimension joins and selectivity filters. The metric is per-query and per-flight latency at a chosen scale factor. It specifically exercises star-join performance, dimension filtering, and aggregation — the operations a dimensional warehouse runs most.

Methodology

SSB reshapes TPC-H's normalized schema into one large fact table (LINEORDER) joined to four dimensions: CUSTOMER, SUPPLIER, PART, and DATE. Data is generated with a dbgen variant at standard TPC-H-style scale factors, so SSB at SF100 is comparable in raw size to TPC-H at SF100. The 13 queries are grouped into four flights that vary the number of joined dimensions and the restrictiveness of filter predicates, isolating how engines handle selective star joins versus broad aggregations. A run loads the data and times each query, typically reporting cold and warm results. The simple, well-understood structure makes it easy to reason about why one engine outperforms another.

How to Interpret Results

Report the scale factor and whether times are cold or warm. Read results flight by flight: flight 1 is highly selective on one dimension, while later flights join more dimensions and aggregate more data, so a strong star-join optimizer and good predicate pushdown show up in the later flights. Because the schema is a clean star, SSB rewards techniques like dimension-key dictionary encoding, bloom-filter join pushdown, and late materialization. Compare engines only at the same scale factor on the same hardware.

Limitations

SSB is intentionally simple, with only four dimensions and uniform data distributions, so it does not capture the complexity, skew, or query variety of TPC-DS. It tests a single clean star and ignores snowflakes, slowly changing dimensions, and semi-structured data. As a derived, self-run benchmark it produces relative numbers, not audited results. Use SSB as a focused, easy-to-interpret measure of star-join and aggregation performance, alongside richer benchmarks for breadth. Because its clean structure makes performance differences easy to explain, SSB is a favorite teaching and diagnostic benchmark for understanding how columnar engines accelerate star joins.