Controlled SQL query evolution for decision support benchmarks

The synthesis of increased global competitiveness and the acceptance of commercially available multi purpose database management systems (DBMS) for decision support applications requires an ever more critical system evaluation and selection to be completed in a progressively short period of time. Designers of standard benchmarks, individual customer benchmarks and system stress tests alike are struggling to mastermind queries that are both representative to the real world and execute in a reasonable time. Additionally, the enriched functionality of every new DBMS release amplifies the complexity of today's decision support systems calling for a novel approach in query generation for benchmarks. This paper proposes a framework of so called query evolution rules that can be applied to typical decision support queries, written in SQL92. Deployed in combination with QGEN2, the query generator developed by the TPC for TPC-DS ?[13], these rules quickly turn a small set of queries into a large set of semantically similar queries for ad-hoc benchmarking purposes or they can be used to generate thousands of queries quickly to stress test optimizers or query execution engines without much user intervention.