Re: NOARCHIVELOG and rollback segments

"John" <jbradshaw777_at_yahoo.com> wrote in message news:f2f59d82.0311051457.66d575ac_at_posting.google.com...
> "Howard J. Rogers" <hjr_at_dizwell.com> wrote in message
news:<3fa94dae$0$3787$afc38c87_at_news.optusnet.com.au>...
>
> > > 2. Can a transaction span more than one rollback segment?
> >
> > No. If it starts in one rollback segment, it must finish in that
rollback
> > segment. In 9i, with "undo segments" (which are just rollback segments
with
> > a different name) that's modified slightly because one segment can
'steal
> > extents' from another, and so a transaction can indeed appear to cross
> > between segments. But that's uniquely 9i.
> >
>
> Thanks, Howard. Speaking of 9i AUM, are you aware of anything problmes
> associated with it?

Yes, one or two. Check out Metalink, because they are all platform-specific. There used to be dozens (we're talking bugs here). I wouldn't have touched it in 9.0.1, but most seem fixed in 9.2. But check.

In terms of whether a (hypothetically) bug-free implementation has intrinsic problems, yes there are still some gotchas. The one I fell into was creating enormous undo tablespaces because flashback seemed such a good idea, and I had the disk space. Unfortunately, having oodles of free space left causes Oracle to put off the day when new transactions start sharing existing undo segments: Oracle tends to create new undo segments for each new transaction until space pressure is felt (it's a complex algorithm, so it's not that 'pat', but it's along those lines). Net result: huge tablespace means lots of undo segments. Each undo segment has a segment header block. Therefore, huge tablespace means lots of undo segment header blocks clogging up my buffer cache.

But, sized appropriately, I actually quite like AUM. But ask Jonathan: he's not so keen on it, I think.

>It's been around for more than a couple of years,

Not really. I seem to recall teaching one of the first 9i courses in Oz around about August last year. So just over a year and a bit. It's still not totally mature, I think.

> I would assume most of the kinks would have been fixed by now.

Assuming anything is always risky! Particularly with AUM. It's pretty close, and I've never encountered any disasters with it, but check Metalink very, very carefully.

>What do
> you think of the old dictionary views like v$rollsats and
> dba_rollback_segs? Some of the stuff in there do not make much sense
> anymore. 'Optimal' for example.

Worth the price of the upgrade on its own, v$undostat is your friend, whether you choose to run AUM or not. Tells you your max concurrency, the longest running query, and out-of-space error counts. Brilliant stuff. The old stuff is still there, true. But I haven't looked at them for a while (v$rollstat is still useful for finding out which specific undo segment has been used to house a transaction's undo, though).

Regards
HJR
>
>
> John
Received on Wed Nov 05 2003 - 17:07:04 CST