Re: Rollback segments

From: Naushi Hussain <naushi.hussain_at_alliedsignal.com>
Date: Tue, 16 Jan 2001 10:49:03 -0700
Message-ID: <3A64898F.2A89A427@alliedsignal.com>

just like to mention here that more than one transactions can stay in one extent if there is room but a block can contain only one transaction at a time.

"Howard J. Rogers" wrote:

> "Greg Stark" <greg-spare-1_at_mit.edu> wrote in message
> news:87k882xiea.fsf_at_stark.yi.org...
> > "Howard J. Rogers" <howardjr_at_www.com> writes:
> >
> > > Incidentally, you are asking for trouble having rollback segments with
 only
> > > two extents. The chances of them having to acquire more extents is very
> > > high (assuming you eventually bring them online).
> > >
> > > Rollbacks should have around half a dozen extents to minimise the need
 to
> > > extend dynamically
> >
> > I don't understand how you can say that without knowing the size of the
> > extents and activity. Is there any major functional difference between a
> > rollback segment with 6 1M extents and a rollback segment with 2 3M
 extents?
> >
> >
> > --
> > greg

>

> Yes, a huge one. New transactions cannot move into an extent of a rollback
> segment which has an active transaction within it. With just two extents,
> the chances of filling up the second extent and thereby needing to move back
> into the first which is likely to still have one active transaction in it
> (however trivial that transaction is) is extremely high. The more extents
> you have, the lower the probability. When it can't move back into the first
> extent because of the presence of an active transaction, the segment will
> instead acquire (dynamically) an additional extent.

>

> Since it is the presence of a single transaction that prevents extent reuse,
> it's the number of extents, not the size of them per se that is the trouble.
> 2 3M extents are, other things being equal, much more likely to have a
> single live transaction in them somewhere than 6 1M extents. Of course, it
> should take longer to fill up the second extent, and thus need (or, rather,
> want) to move back into the first... but we are talking probabilities here.
> You are more *likely* to need to acquire a third extent. Naturaly, activity
> levels determine the actual need, or lack thereof, for doing so.

>

> Regards
> HJR
Received on Tue Jan 16 2001 - 11:49:03 CST