Thanks Dieter Carrie blog is very useful and having date columns not samples is on our plan now.
I've noticed when minimizing the Sample size the number of values computed decreases; which can be expected.
My Concern about the point you mentioned it will always gets the first % blocks specified in the sample which can be the first / last blocks inserted.
in case of highly unique columns ( only 10 values for the column for the column); a possibly bad plan may occur; once a specific value of these 10 values is missed in the samples stats and used in a predicate;
which what's the trade-off for accuracy / stats time is all about.
My question : is there any way we can change the way the sample is collected other then picking the first blocks; like random blocks for the same percentage ?
↧