CUBE, ROLLUP, and OUTER/ANTI/SEMI-join speedups


This question came from @christoph on another channel:

was very happy to read today’s announcement of v6.8 becoming available.

As we already have had JIT enabled, this is no change to us.

However we are excited about the improved GROUP BY, CUBE and JOIN performances as stated here:

Do you have any figures explaining how much faster GROUP BYs and CUBEs are?

Also, the website states that v6.8 " speeds up execution of some queries" in regards to JOINs. Can you elaborate which JOINs are faster?

We usually only do (with L being a large fact table and S being a small dimension table)

  • L left join S
  • L join S


Re: CUBE and ROLLUP, query 67 from TPC-DS speedup 5.5X for us.

Re: Join speed improvement, some right and left, anti-, semi-, and outer-joins are faster. TPC-H query 21 speeds up by 4.3X in one of our internal runs.

I can’t give you any more general description of the improvement than that. It depends on your data and queries.