Re: Raw Devices: Increased Performance?

From: Glenn <glennf_at_crg8.sequent.com>
Date: 1996/07/19
Message-ID: <4soes8$7fs_at_scel.sequent.com>#1/1


Once a block is put on the MRU end of the chain, it will follow the normal algorithms and rows that are accesed most often will tend to stay in memory. The bigger issue is your first statment:

"What happens if you've not got enough space the SGA to hold the complete table?"

If you haven't got space to hold the complete table in Memory and you are performing full table scans, then there is not much reason to alter the table to CACHE. This is done, then your entire SGA will be abolished of usable every time you perform a scan. Additionally, the portion of the table that would remain in the SGA would not be used on the next full table scan and you would effectivly read the entire table from disk anyway.

In article <4snl9t$gtj_at_atlas.aethos.co.uk>, Steve Holdoway <steve_at_aethos.demon.co.uk> wrote:
>glennf_at_crg8.sequent.com (Glenn "Daddy G" Fawcett) wrote:
>
>>In Oracle 7.3 you can alter a specific table's properties to place buffers
>>obtained by a tablescan at the MRU end of the LRU chain.
 

>> ALTER TABLE abc123 CACHE;
>
>What happens if you've not got enough space in the SGA to hold the
>complete table? What I'm praying that you'll say is that the most
>commonly used parts of the table are kept in memory. Or can you force
>this to happen if it's not the default case??

-- 
Glenn Fawcett,  Technical Marketing Engineer         voice:    (503)578-3712
Sequent Computer Systems                               FAX:    (503)578-5453
15450 SW Koll Pkwy (MS CHA1-371)               UUCP:     ..uunet!sequent!glennf
Beaverton, Oregon 97006-6063               internet:         glennf_at_sequent.com
Received on Fri Jul 19 1996 - 00:00:00 CEST

Original text of this message