Re: the IN club: Oracle "unpublished" information

I have a few thoughts,

First I don't believe that wanting to know better how the Oracle software behaves (distinct from works) is at all a bad thing for either a DBA or a Developer to wish for themselves, so no I don't think that you are off your rocker, or at least if you are, the company of fools in which you find yourself is large and distinguished. In principle if we understand how Oracle behaves we will be able to reliably predict how our applications will behave and what the experience of our users will be like. Ok, so maybe I'm a little mad as well.

Sometimes however understanding how Oracle behaves will require understanding how Oracle works - the implementation detail. The CBO example that you have referred to looks like a great example of this. Here though you are likely to find that this is dependent upon, your version primarily; your setup (try running Jonathan's repeatable examples with ASSM in place for example). Expecting people who have a job of keeping production systems up and running for paying customers (and answering the phone!) to keep up to date with all that is somewhat ambitious.

In addition, Oracle is not an OpenSource company, they do have internal intellectual property to protect. It's likely support can't even see this (I don't know but I can't imagine for example that a support contact could see how the HASH JOIN algorithmn works in 10.2.0.3 with systems stats in place, histograms on join columns and so on, still less that theywould see how that was different from 9.2.0.1). Your stance on the desirability of that may vary according to what you think about IP and software, but it's still going to be the case.

Finally, if you can, and it looks like you did, produce a simple straightforward test case that exhibits obviously bad behaviour, you can certainly push for a bug to be logged and a workaround to be found (which is actually what your customers are likely to want). This is most definitely within the capacity of support.

On 3/14/07, Charles Schultz <sacrophyte_at_gmail.com> wrote:
> Good day, all,
>
> No, I am not offering any inside or exclusive information. I wish I could.
> =)
>
> Rather, I am curious about many things Oracle, and in particular, how do
you
> folks go about talking nicely with Oracle Developers and decision-makers?
I
> am a bit frustrated by Oracle Support's obdurate stance to maintain
> "internal" information as "classified" or "unpublished". In my quest to
> learn more about How Oracle Works, I usually end up bugging people on this
> list because Oracle Support refuses to answer my questions (except on rare
> occasions). Can one join a well-respected OUG and gain access to some
> players? What other resources are there?
>
> For instance, I was recently trading emails with Wolfgang Breitling, and
he
> was patiently clarifying his paper "Joins, Skews and Histograms" (with
kudos
> to Alberto Dell'Era). For an extremely simple case (join a 20 row table to
a
> 40 row table where half of the rows from the first are repeated x4 in the
> latter), Wolfgang (and Alberto) demonstrate that Oracle uses a highly
> complex algorithm to calculate cardinality only when frequency histograms
> are used (as opposed to SIZE AUTO or SIZE SKEW, or no stats). The kicker
is
> that the estimated cardinality is really bad (45% error). I filed an SR
with
> Oracle to ask "why" and was given the standard boilerplate.
>
> Granted, I still have a long way to go before anyone would even consider
me
> expert material, but it does not take a genius to scratch his head and go
> "huh?". I can imagine that Oracle justifies such a complex algorithm
because
> they want to be able to use it consistently for complex situations. But
why
> for such a simple case?
>
> My hat is off to folks like Alberto who apply their significant intellect
> and experience to the enigma of "what the heck is going on." Is there any
> recourse to gently nudge Oracle Development towards a simpler solution for
a
> simple case? Or am I totally off my rocker and just way too naive to even
be
> in this discussion?
>
> --
> Charles Schultz

-- 
Niall Litchfield
Oracle DBA
http://www.orawin.info

--
http://www.freelists.org/webpage/oracle-l