Comments on: Selectivity Guesses in absence of Statistics

By: joe

joe — Mon, 21 May 2012 07:06:59 +0000

Hi Chandan,

If auto-create / auto-update statistics are enabled (which most often they should be) – the guesses don’t come in to play (for this scenario at least). Basically the guesses come into play when there are no stats involved (whether stats objects or stats associated with an index).

Cheers

By: Chandan

Chandan — Sun, 20 May 2012 22:05:28 +0000

Thanks for writing this Joe. Does it mean that in a table which is a heap and with no -nonclustered index, the actual row-counts guessed by the optimizer while compiling a plan will always be wrong?

What about a heap with some non-clustered index on some columns?

Thanks
Chandan

By: tobi

tobi — Thu, 10 May 2012 21:22:17 +0000

This table could be used to construct "cardinality hints" by introducing useless predicates which alter misestimated cardinality. A nasty hack, but it seems nicer that join hints which cause side-effects like forcing join order.