Re: Oracle's Myth: keep tables and indexes in separate tablespaces

From: steve <me_at_me.com>
Date: Sun, 12 Oct 2003 6:57:42 +0800
Message-ID: <0001HW.BBAEABE60002D4E9F0284600@news.newsguy.com>

On Sat, 11 Oct 2003 19:51:40 +0800, Niall Litchfield wrote (in message <3f87eeb5$0$264$cc9e4d1f_at_news.dial.pipex.com>):

> "steve" <me_at_me.com> wrote in message
> news:0001HW.BBADE6080002399FF0284600_at_news.newsguy.com...

>> I think we  all need to be a little clearer on what we are saying.
>> 
>> when we are saying separate tables and indexes, do we mean:
>> 
>> 1. in separate segmnents on the same disk.
>> 2. in sequential disk blocks within the same oracle segment
>> 3. on seperate disks on the same controller.
>> 4. on separate disks on different controllers.

>
> Um well as tables and indexes are both types of segment 1 and 2 are
> nonsensical statements. Did you perhaps mean tablespace or
> something else instead of segment?
>
> The myth is that just by putting data and indexes in seperate tablespaces
> performance improves. This is demonstrably untrue. Putting contending
> segments (which tables and their indexes may be) on different disks does
> improve performance but this is not the same thing.
>

Sure if it is a Global statment that performance ALWAYS improves by separating them out then it can be proven very easily that it is nonsence.

sorry segment=tablespace.
1. i was refering to the oracle paper where they state that trying to put the data on a certain physical part of the disc can improve the performance. 2. even with an oracle tablespace, it is not possible to gurantee that the rawdata is in sequential blocks( at the disk level, not the table level), so in theory you can have 2 identical setups ( hardware wise) that would perform differently, if one of them had tablespace raw areas that were fragmented at the raw level. Received on Sat Oct 11 2003 - 17:57:42 CDT