Re: Index management

Mike Ault wrote:

> Howard,
>
> Your graciousness only eclipses your tactfulness. At sites with tables
> that contain millions (in one or two cases, billions) of rows doing a
> compute rather than estimate on statistics is not realistic or
> possible. Had you the experience that your 30 years in the field
> should have granted you this would be obvious.

30 years? That would mean I was 10 when I started. Slightly ambitious, I would think.

But whatever. You just missed the point, Mike. Again. I wasn't suggesting doing a compute statistics on a billion row table. I was, however, suggesting that if you are about to draw conclusions from a situation with which to infect the mind of Oracle users for years to come, you might just make sure that your statistics are spot on before doing so.

Why not just start out testing a hypothesis on a, say, 4000-row table. Then you can move to testing on a 40,000-row table. And if your hypothesis is still holding up at that point, try a 4,000,000-row table. Just before you're ready to publish, find a client with a billion-row table you can test it out on. Just a suggestion, of course.

The gist of that technique, you know, is that one has an Oracle database at *home* on which to test things out. I actually test stuff all the way up to a 400 million row table, at home. I start my test, go to bed, and assess the results the next morning. As I say, it's just a suggestion. But then *I* don't have a problem with *calculating* statistics for a 400 million row table.

> You also assume that the Oracle system is bug free and always does
> everything correctly. I can't afford to assume that as my experience
> has shown otherwise. Shoot, they can't even get simple row counts
> correct much of the time.

Strawman. Again. I make no such assumptions, so your comments are unwarranted.

>
> I understand what the absolute clustering factor

Mike, will you do me, yourself and (most importantly) the *entire* *Oracle* *community* a really big favour, and stop making up words and phrases that sound really big and important, but which actually mean nothing? Otherwise people will start to think it's just a lot of pompous hot air designed to cover a paucity of actual content.

We weren't talking about an "absolute clustering factor". Just the clustering factor that's visible in DBA_INDEXES as one, nice, simple number. It doesn't need any extra adjectives.

> is, and also
> understand that there are no absolutes in the real world, everything
> must be tested

WELL LEARN HOW TO TEST PROPERLY THEN!!! You will save us all from errors and mistakes and myths promulgated, not by Oracle bugs or coding mistakes, but by Mike Ault, author, who is in the privileged position of being able to publish and widely disemminate his views and opinions... and therefore has a greater responsibility than most to make sure those views and opinions are grounded in solid fact and even more solid analytical and testing technique.

> as algorithms are implemented incorrectly on a daily
> basis across much of the industry.

This thread has had nothing whatsoever to do with incorrect implementation of an algorithm. It has had everything to do with someone who claimed to know something finally admitting that what he thought he knew, he'd never tested properly; and grudgingly admitting that he was therefore wrong.

As I said, and I meant it, that is good news for the Oracle user community. Knowledge 1, ignorance-myth 0.

What I didn't say, because I was trying to be nice about it, is that the only reason ignorance and myth were ever in the game in the first place is because of people like you, Mike, drawing sweeping and erroneous conclusions from, it would now appear, not much actual data and a worse testing technique.

>
> As far as I am concerned, the tread is closed.

I agree. You've admitted you were wrong, and anything else is just superfluous. The Oracle community's store of knowledge and fact just went up by 1. I think we can all agree on that.

HJR
>
> Mike
Received on Tue May 11 2004 - 15:03:35 CDT