Dave,
during a 'db file sequential read', an index is _not_ accessed sequentially.
An index is not a sequential structure, so reading from an index in order
will cause multiple seeks on the index itself. And we're talking single user
here
regards,
Hans
Reply-To: [EMAIL PROTECTED]
To:
Hans,
Your statement is true except in the case of a fast full-index scan.
But that's not my point. What I'm trying to say is:
1. In scenarios where response time is important, for example when you
want to obtain the first n rows of a query result as quickly as
possible, then access time
.
Jared
Dave Hau [EMAIL PROTECTED]
Sent by: [EMAIL PROTECTED]
10/08/2003 04:19 PM
Please respond to ORACLE-L
To:Multiple recipients of list ORACLE-L
[EMAIL PROTECTED]
cc:
Subject:Re: SAME and separating disk and index tablespaces
Hi Gaja
- Original Message -
the striped array. However, this does not improve access time. If you
have your tables and indexes on the same striped array, necessarily the
two I/O's have to be done sequentially, incurring two times access time
at a minimum. However, if you separate the
Title: RE: SAME and separating disk and index tablespaces
Maybe we have been lucky. But we use the SAME methodology. We have removed a considerable amount of human effort in regards to layout of datafiles and disk layout. And based on the stats that I have seen from the Storage team/SA's, we
Gaja Krishna Vaidyanatha scribbled on the wall in glitter crayon:
Hi Hans/Vikas,
I tend to agree that the old draconian rule that thou
shalt always separate indexes from tables may not
apply any more. We used to apply that principle in the
past when the number of available spindles was
Title: RE: SAME and separating disk and index tablespaces
Good stuff. Plus, watch this:
If saving dozens of hours of labor cost
actually does cost a full 5% performance penalty on access time, and if reads
from disk account for 10% of total response time for a given user action
David and list,
Some points to keep in mind in our discussion:
* Throughput can be measured both by IOPS and MB./sec,
it depends on the application, whether it is purely
transactional or is just hauling loads of data.
Access time normally is a function of the number of
I/O operations the disk
Gaja,
No cups in this analogy?
I'm thoroughly dissappointed.
I would have liked the idea of a twelve pack that can only provide one can at a time, vs. 2 six packs that could each provide a beer.
Paul
Gaja Krishna Vaidyanatha [EMAIL PROTECTED] wrote:
David and list,I think this provides
Vikas,
The answer is an enthusiastic yes. This is purely from
an administrative and manageability standpoint. For
example, if you have INDEX and DATA segments separated
in 2 different tablespaces, the backup of these
tablespaces can be done INDEPENDENTLY. This is
relevant, as if you were to
Thanks Gaja ! Does it also make sense from a performance perspective
(I/O issues due to concurrent access of index and data ) to separate
them or is that point moot once you apply the SAME methodology ?
-Original Message-
Gaja Krishna Vaidyanatha
Sent: Wednesday, October 08, 2003 9:24
Vikas,
Spend an hour on reading this usenet thread:
http://groups.google.nl/groups?hl=nllr=ie=UTF-8oe=UTF-8threadm=brjz8.15%24707.245%40news.oracle.comrnum=1prev=/groups%3Fhl%3Dnl%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26q%3Drogers%2Bseparate%2Bdata%2Bindex
It will open your eyes about separating
Hi Hans/Vikas,
I tend to agree that the old draconian rule that thou
shalt always separate indexes from tables may not
apply any more. We used to apply that principle in the
past when the number of available spindles was not
adequate. Seems like with 256G drives in the market,
we are being pushed
I completely agree with Howard about one thing: indexes and data don't
necessarily compete against each other for I/O capacity.
So the following is *not* true: You should separate indexes and data
into different tablespaces because they compete so strenuously for I/O
capacity.
However, the
Hi Gaja,
I agree that throughput can always be improved by adding more drives to
the striped array. However, this does not improve access time. If you
have your tables and indexes on the same striped array, necessarily the
two I/O's have to be done sequentially, incurring two times access
Great responses ! Thanks very much ..
-Original Message-
Dave Hau
Sent: Wednesday, October 08, 2003 3:19 PM
To: Multiple recipients of list ORACLE-L
Hi Gaja,
I agree that throughput can always be improved by adding more drives to
the striped array. However, this does not improve
-L [EMAIL PROTECTED]
cc:
Subject:Re: SAME and separating disk and index tablespaces
Hi Gaja,
I agree that throughput can always be improved by adding more drives to
the striped array. However, this does not improve access time. If you
have your tables and indexes on the same
17 matches
Mail list logo
