Re: SECUREFILES disaster
Date: Sat, 28 Apr 2012 16:35:51 +1000
Message-ID: <jng347$l32$1_at_dont-email.me>
Jonathan Lewis wrote,on my timestamp of 28/04/2012 5:54 AM:
> | I share this experience as well. A while ago I was
> | involved in a db design that required updates to a LOB.
> | We ended up simply inserting a new version and deleting the
> | old one: it was much faster than waiting for Oracle's
> | "efficient" handling of LOB updates...
>
> That's odd, because if you haven't written code to do page-based updates to
> LOBs, Oracle deletes the old one and inserts a new one when you do an
> update.
Like I said - a while ago, or more specifically: 9ir2. It also had the habit (still there, I believe?) of using the LOB tablespace as the UNDO tablespace for the operation. Which made sense, since copying the original to a separate UNDO would have been a big hit on performance. But the clincher with us was with all the indexes on the row and which UNDO they used. It was much faster to just delete the row and insert a new one than wait for Oracle to figure out how to create a "hole" or fill one in the LOB tablespace and work out how/where to write to the system UNDO. Received on Sat Apr 28 2012 - 01:35:51 CDT