The main benchmarks that these compilers compete on are those from http://www.tpc.org/ http://www.tpc.org/. They are synthetic datasets which are fairly naively

The main benchmarks that these compilers compete on are those from http://www.tpc.org/ http://www.tpc.org/. They are synthetic datasets which are fairly naively generated. http://www.vldb.org/pvldb/vol9/p204-leis.pdf http://www.vldb.org/pvldb/vol9/p204-leis.pdf is a really interesting paper that looks at how some of the heuristics used in those compilers work well on TPC but break down on real datasets. It can give you a good intuition of when you have to be wary of bad plans.