Dmitry Yemanov wrote:

>As you understand, effective query optimization can be achieved only with an
>adequate cost calculated. The two major criterias for the cost estimation
>are cardinality and selectivity. The engine must support all required stored
>statistics which can be useful to properly estimate these values. Currently
>the only calculated statistics is the index selectivity, extended in FB2 to
>support per-segment selectivities. This proposal describes which other types
>of statistics can be used, how they should be stored and possible update
>policies.
>
>
>

Before you get deep in the implementation, may I suggest you try and
test some mockups? It would be nice to know by experimental evidence
rather intuitive what statistics are actually going to help in making
correct decisions. I have no doubt that better data would make for more
accurate estimates, but the best estimate are going to continue to be
wildly inaccurate. It may be that finer tools may be necessary, or some
formula that can deal with uneven distribution. I may be all wrong, but
I'm not convinced that better cardinality and simple selectivity values
is going to help all that much. Convince me with evidence.


--

Jim Starkey
Netfrastructure, Inc.
978 526-1376