Vladimir,

I feel the same way. MVs are more powerful and general, and with some effort 
they could be just as efficient as other approaches. 

One problem that needs to be solved is the “registration problem”. If you have 
a lot of MVs they all have to be registered in the planner’s search space, 
otherwise the planner will not find them.

I would like to see a benchmark for this. Say we have 1,000 MVs and we plan 
10,000 queries that potentially use them. The goal is that the planning cost Is 
comparable to only having 1 MV. We could achieve the goal by re-using the MVs 
across planner instances (thereby amortizing the cost of creating them) and 
“index” the planner’s search space.

The Lattice structure essentially does this for families for join-aggregate MVs 
(summary tables). We need something similar for project-sort MVs (indexes). 

Julian

> On Jun 2, 2020, at 12:05 AM, Vladimir Ozerov <[email protected]> wrote:
> 

Reply via email to