We need a re-run of DBT-3 benchmark with the new features:
Engine-independent statistics in MariaDB 10.0 (note: Timour will provide a 10.0-based tree with latest fixes to EITS and subquery optimizations)
Innodb's persistent statistics in MySQL-5.6
The primary goal is to explore what query plans will be used with Engine-independent table statistics (further called EITS).
EITS are persistent, predictable (statistics calculation has no randomness elements), and precise. The statistics need to be collected manually, and there is an optimizer setting to get them to be used (see https://kb.askmonty.org/en/engine-independent-table-statistics/ for details).
MySQL-5.6 and their persistent statistics should be used as something to compare against. I suppose, increasing innodb_stats_persistent_sample_pages should reduce the number of chosen plans (to one plan even?) ?
The benchmark should be a DBT-3 run. It should be IO-bound.