TPCx-BB (BigBench)

TPCx-BB, also known as BigBench, is an express big-data analytics benchmark from the Transaction Processing Performance Council. It targets distributed analytics platforms such as Hadoop, Spark, and big-data SQL engines, measuring an end-to-end workload that blends structured, semi-structured, and unstructured data processing. It was created because traditional warehouse benchmarks did not capture the variety of modern big-data pipelines.

What It Measures

The primary metric is BBQpm@SF (BigBench queries per minute at a scale factor), plus price/performance and energy options. The workload comprises 30 queries against a retail data set covering structured sales tables, semi-structured web clickstreams, and unstructured product reviews. The queries exercise pure SQL, machine-learning tasks (such as clustering and classification), and natural-language processing (such as sentiment analysis), so the benchmark measures a platform's breadth, not just SQL speed.

Methodology

TPCx-BB extends a TPC-DS-like retail schema with additional semi-structured and unstructured sources. Data is generated at scale factors expressed in gigabytes, scaling into the multi-terabyte range. As an "express" benchmark, it ships as a runnable kit so vendors execute a standardized implementation rather than building their own. A compliant run measures data loading, a power test (single query stream), and throughput tests (concurrent streams), then combines them into BBQpm. The 30 queries are distributed across the three processing styles, requiring the platform to integrate SQL engines with ML libraries and text processing in one pipeline. Official results are audited.

How to Interpret Results

Report the scale factor and the number of concurrent streams. Because the workload spans SQL, ML, and NLP, a strong result reflects an integrated platform rather than a fast SQL engine alone — examine which query categories dominate the runtime. Price/performance and, where reported, energy efficiency matter for cluster sizing decisions. As with other TPC benchmarks, prefer audited results and verify the platform version and node count in the disclosure report.

Limitations

The mixed workload makes results harder to compare across very different platforms, and the ML/NLP components depend on library choices that can shift outcomes. Full runs require a substantial cluster and are operationally heavy, so published results are relatively few. The retail-analytics domain may not match other big-data use cases such as IoT or graph processing. Use TPCx-BB to assess an integrated analytics platform's end-to-end capability rather than a single engine's raw speed. As one of the few standardized benchmarks that spans SQL, machine learning, and text processing in a single workload, it remains a useful gauge of how unified a big-data platform truly is.