Re: Row migrated in new data blocks

Jim Kennedy schreef:

> "Fabio Cannavaro" <fatti_at_icazzi.tuoi> wrote in message
> news:44e46322$0$30239$4fafbaef_at_reader1.news.tin.it...
> > If a row in data block, after some heavy update, is too long and it is
> > migrated to a new data block (assuming it can contain it) I read that
> > original piece row is preserved from Oracle to point to new row that
> > also take the original rowid.
> >
> > Now I think that in some cases database should be veru full of "row
> > pointing" rows...and I imagine this is not so good...
> >
> > Is it so frequent?
> >
> > Is there a way to optimize this situation?
> >
> > Regards.
> >
> > F.
> Yes, it is a bad thing if widespread. It effectively doubles your IO. Is
> it frequent? It depends. To prevent it you need to adjust the pctfree and
> pctused. (lower the pctused of the table so updates will fit.)
> Jim

someone is mixing up row migration and row chaining here. The OP decribes migration, as I read it. I have never seen performance tests examining this, so, be our guest.
Chaining is when a row doesn't "fit" and thus uses two (or more) blocks.
Migrated rows will not affect full table scans, it will affect indexed reads.

See asktom.oracle.com for some nice demo and explanation Received on Thu Aug 17 2006 - 09:58:42 CDT