Re: Parameter to influence Oracle's Idea of IO Cost?

From: EscVector <Junk_at_webthere.com>
Date: 9 May 2007 11:03:07 -0700
Message-ID: <1178733787.750908.297980@n59g2000hsh.googlegroups.com>

> from values of 0 and 100 for those but oltp workloads might find 90
> and 25 more appropriate ( perhaps ).

Again generics...
What about walking the parameter settings from for optimizer_index_cost_adj from 1-10000.
Strange things happen at different numbers. A plan might be good at 42 but then bad at 43 and good again at 2000.

It's all about the plan and the response time. Mr. Kyte has many generics in his stuff, but he counters this by testing..... Don't use Tom's or anyone else's numbers. Test for yourself.

What about permutations:

TABLE_NAME                     TABLE_TYPE  COMMENTS
------------------------------ -----------
-----------------------------------------------------------------------------------------
OPT1                           TABLE       The range of values is 1 to
10000
OPT2                           TABLE       The range of values 0 to
100
OPT3                           TABLE       possible values on in test
laptop 2,4,8,16,32,64,128
OPT4                           TABLE       possible values - all_rows,

first_rows, first_rows_n. n can equal 1,10,100,1000

Take the above settings and the possible permutations is:

42,420,000

That's a lot to test :)

So how do we apply generics to the possible 42,420,000 parameter combinations represented.

I don't like generics...Everything is singular...so the only way to know is to characterize the workload, but know that every change creates a new system. Received on Wed May 09 2007 - 13:03:07 CDT