基准测试，通过运行一系列具有代表性的程序来评价某个系统，其结果可以与其它系统的测试结果相比较，从而将不同系统按照性能高低排序。在评估系统性能和其它某些方面，基准测试具有重要的作用。然而，运行基准测试是一件十分耗时的工作。为此，我们通过缩小基准的测试集以减少测试时间来缓解这个问题。 本文主要的研究对象是由TPC发布的决策支持基准——TPC-H，通过TPC官网上公布的TPC-H测试数据，按照执行时间将22个查询分为5类，从每类中分别选择一条语句作为该类代表，组成一个只含有5个查询的子集，以达到缩短测试所需时间的目的，同时在评估系统时保证80%以上的正确率。最后我们利用dbt3测试工具对PostgreSQL进行测试，以验证上述子集是否能达到以上效果，并且得出结论。 Benchmarking , which evaluate a computer system with a set of representative programs, can rank it against other systems by comparing their test results. In evaluating the performance of systems and others, benchmarking plays an important role. However, benchmarking is a time-consuming task. So, we alleviate this problem in the way of reducing the benchmark suite to reduce the execution time. In this paper, we research in TPC-H, a decision support system benchmark published by TPC. By using the data in TPC official website, we classify the 22 original queries into 5 classes according to the execution time. Then choose one query from every class as the representative. The subset of only 5 queries achieve a reduction of the execution time and is able to rank different systems with more than 80% accuracy in comparison with the original order. Finally, we test PostgreSQL with dbt3 in order to prove whether the subset does have that effect，and draw a conclusion.