RE: Oracle Book Mal-practice...
Date: Fri, 22 May 2009 12:30:07 -0400
Well, to be clear, while Oracle does indeed use the rowids to access the actual table, the idea behind separating them onto separate raid/disk groups is that the enemy of traditional/old-school storage arrays is disk seeks. The idea being that if the index+table are located on the same disk, there's seek+index read+seek+table as a linear operation. If they're on different disks, you may still have two seeks, but perhaps the head will already be over the table on the second disk, and you can pipeline more I/Os, as my index disk can already be servicing more index read I/Os while the table disk is handling that.
All that being said, there's so many layers of abstraction today already, and more on the way, that trying to optimize like that is futile, and almost certainly not worth it. Between volume managers, dynamic reallocation and tiering, deduplication, solid-state drives, multi-stage array caches, etc., it's impossible to make generalizations anymore.
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Mercadante, Thomas F
Sent: Friday, May 22, 2009 12:18 PM
To: tomdaytwo_at_gmail.com; Oracle-L
Subject: RE: Oracle Book Mal-practice...
I agree with your statement about SAN and tablespaces. To me, it is getting more and more irrelevant. I depend on the SAN administrator to spread the data around on the drives.
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Thomas Day
Sent: Friday, May 22, 2009 11:04 AM
Subject: Re: Oracle Book Mal-practice...
I was having a discussion with a junior DBA the other day about whether tables and indexes need to be in separate tablespaces and the issue of concurrent access came up.
My position is that Oracle always reads the index and the uses the rowid(s) to access the table. There is no issue of concurrent access. However, she pulled out the latest and greatest Oracle 11 book and sure enough the author repeated the old myth about concurrent access and the need to separate indexes and tables.
How can you fight this? With SANs and logical disks there's no certainty that separate tablespaces means that you're using separate read/write heads. I'm getting as tired of this argument as I am of the RAID5 argument. It shouldn't even be a point of discussion.
Doesn't Oracle have a vested interest in seeing that books about Oracle have correct information or does that just make for more opportunities for Oracle Consulting?Received on Fri May 22 2009 - 11:30:07 CDT