Re: Query Slow in 10g, runs very fast in 8i

Niall Litchfield wrote:
> EscVector wrote:
> > Jonathan Lewis wrote:
> <snip>
> > > What's the rationale behind that suggestion ?
> >
> >
> > I think you said it best: "The first (and only) rule of optimization
> > is: "Avoid unnecessary effort".
>
> There are of course 3 words in that sentence and the middle one is
> rather important. It isn't a synonym for "make no effort". Or in the
> words of the scientist "Make it as simple as possible, *but no
> simpler*"
>
> > This no brain approach avoids people effort. If it doesn't work, add
> > brain.
>
> If I wanted to avoid human effort I'd throw machine effort at it, in
> the form of the Oracle provided tuning tools or from a third party.
> Changing parameters because once upon a time they helped unrelated
> problems is not really the most appropriate approach that I have ever
> seen. I've twice had to visit my doctors this year. They have about 5
> minutes to diagnose what is up with me. I'd be mortified if they merely
> prescribed something because it had helped someone else on another
> occasion - especially if it waa something that made drastic alterations
> to my biological state.
>
> > The other simple answer is I've had it work in the past when we had to
> > make very quick adjustments to things we couldn't test or spend time
> > researching and that couldn't be changed.
>
> Why change anything if you can't test it. If the org can't afford to
> make an informed decision why make any change at all? After all it
> isn't worth the cost of testing.
>
> > This no brain approach avoids people effort which is usually the most
> > costly component to any effort. To something while your doing the real
> > thing. That's the rational.
>
> Isn't it more sensible to do nothing while you are doing the real
> thing? I mean the suggestion you make *could* be an extraordinarily bad
> suggestion reducing elapsed time from 11 hours to 11 days...
>
> Niall

"If the org can't afford to make an informed decision why make any change at all? "

There are many, many situations in which the decision must be made regardless. It has nothing to do with not being able to afford the decision and it may not be mine to make, but mine to do. The key is get it done now. If 25% of your income were depending on meeting a failing SLA, would you wait and analyze if analysis would fail your SLA, or would you attempt something, or would you sit and do nothing. Do or die.

If they are having problems like this and the cause is unknown, the skill level to which they are likely to attain in the next few days is not that of expert. If you have the time and no harm done, then why not attempt the unknown if it takes a few seconds? I fail to see the harm.

Expert analysis takes time and more importantly, planning. I threw this maverick suggestion out there to generate controversy. If the best experts in the world were called to fix a problem, billable by the hour, and they said that fixing the problem would require a 10046 trace that would take 11 days, what is in "afford". What if the manager, who knows nothing, tinkers with opt settings and solves the problem from out from underneath the experts in the first 3 hours while they were analyzing a 10k line 10053 trace. In this situation, the experts, don't get paid, so how's that for affordable. As I said, there are many, many situations. I'm just throwing this one out there.

I'd probably not throw machine effort at it. I'd use my experience to make a educated guess if necessary based on immediately available data.  The machine has already gotten it wrong. :)

Niall, I appreciate your time and your response. I agree with you 99%.

Two books I like:
http://www.gladwell.com/blink/
http://www.onintelligence.org/ Received on Fri Jan 05 2007 - 21:21:20 CST