Re: Segment management auto clause

"Richard Foote" <richard.foote_at_bigpond.com> wrote in message
>
> Hi Howard,
>
> Point taken.
>
> But you do admit to having taken to a position rather strongly that "you
> would be one mad lunatic" if you would ever use ASSM, except with RAC or
> heavily inserted situations (like you again suggested to the OP) . I don't
> ever recalling you mentioning that ASSM may be useful in other scenarios.

This isn't a matter of "positions". I'm not defending the Western Front here. Either ASSM has overheads, or it doesn't. If it does, then employing it without a counter-balancing good excuse (such as free list contention) seems awfully daft.

You say there is little or no overhead. I say there is. That's all. And if it was just me saying there were overheads, I'd think twice about it. But it isn't. Have a chat with Jonathan, please. In any case: This isn't a religious war where one has to cling to beliefs regardless of the workings of a rational mind.

>
> I guess the point I would make is that it may not be so black and white
...

Oracle Corporation, in all its objective and thoughtful majesty, is currently doing the 'White' rounds. I don't think the white case needs making. What is sorely needed is someone who can objectify the reasons why it isn't as white as they make it appear. And if that happens to make me look like the Arch Advocate of Black, so be it.

But in any event, my testing means that I believe employing ASSM would require the addition of considerable megabytes of Buffer Cache to compensate, the addition of potentially gigabytes of index space to avoid the costs of full scans, and a frontal lobotomy to make one believe that something that was clearly invented for RAC has any meaningful place in an extremely non-RAC situation (I'll buy that a non-RAC, heavy concurrent small inserts, multiple CPUs, might also benefit from it).

Incidentally, had it not been for one or other of my esteemed colleagues complaining to Oracle about my website, my full set of test results would currently be available for inspection and rebuttal. But I can't publish what isn't mine, so if I have to remain silent on the details, it nevertheless remains the case that in test after test, ASSM has provided either negligible advantage, no advantage whatsoever, or significant disadvantage... so I know where I believe the balance to be.

>
> >
> > >
> > > Secondly, there was some debate regarding each extent requiring it's
own
> > set
> > > of bitmap blocks. This is incorrect. One bitmap block can map to many
> > > extents and indeed as an object grows, Oracle assigns more bitmaps
> within
> > a
> > > bitmap block. Basically the ratio of bitmap blocks is based on a
sliding
> > > decreasing proportion to the number of blocks (not extents) in a
segment
> > > (show anyone interested a few block dumps if you like)
> >
> > Don't need to to me at any rate. It isn't one-bitmap-for-one-extent, I
> don't
> > think I suggested with any absolute certainty that it *was* one-for-one,
> but
> > I haven't been saying anything of late other than that it starts
> > one-for-one.
>
> But that's incorrect.
>
> If the extents are tiny, then one first level bitmap blocks could span
> several extents.
>
> If the extents are large, you could have many bitmap blocks ...
>
> Even at the start ...
>
> Cheers
>
> Richard
>
>
Received on Wed May 14 2003 - 23:46:14 CDT