RE: stipe size
- From: "Cary Millsap" <cary.millsap@xxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Wed, 25 Feb 2004 23:03:18 -0600
Watch out, though. If you set the stripe size to *only* the value of the
max atomic I/O size, you've got to make sure that your logical and
physical stripes are aligned. (See Hank Tullis's article at hotsos.com.)
Otherwise, you'll end up having a single I/O call split into two simply
because there'll be a physical boundary in the midst of a logical
stripe.
Cary Millsap
Hotsos Enterprises, Ltd.
http://www.hotsos.com
* Nullius in verba *
Upcoming events:
- Performance Diagnosis 101: 2/24 San Diego, 3/23 Park City, 4/6 Seattle
- Hotsos Symposium 2004: March 7-10 Dallas
- Visit www.hotsos.com for schedule details...
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Mladen Gogala
Sent: Wednesday, February 25, 2004 11:00 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: Re: stipe size
On 02/25/2004 11:31:35 PM, Cary Millsap wrote:
> What is "small"? I think anything less than about 1MB is small, even
for
> TP. Since 7.3, the Oracle kernel does its very best to do large read
> calls. This is a Good Thing. Using stripe sizes that cut a single
large
> I/O call into multiple parts is a Bad Thing on a many-user system.
Well, here's the problem: Unix system can do atomic I/O as large as
the maximum size of it's kio buffer (kernel I/O buffer). On most
systems, kernel I/O buffer max size is between 256k (Solaris 7) and
1M (HP-UX 11, 64 bit). If the stripe size is larger then that,
OS will need 2 or more I/O requests to read a single stripe. That is
the same reasoning as behind not setting DB_FILE_MULTIBLOCK_READ_COUNT
too high. So, to be on the safe side, I'd find out the max size of the
atomic I/O and set stripe size to that value. I think that 256k is just
about ideal. Of course, there is an additional element of RAID on board
cache which is really the element that communicates with the server.
If cache is big, then you want big prefetch, to speed up subsequent
I/O requests. In that case, it's "the bigger, the better" thing
with stripe size.
>
> <what-would-Mladen-do>
> ...However, if the original question is really about "stipe size," I
> believe that Michael Stipe is a little bitty devil, probably not more
> than about 5'4" at most. But this is really outside of my subject of
> expertise.
> </what-would-Mladen-do>
Actually, "Stipe" is a name in my country of origin (Croatia). It is
a derivative of "Stephen" and funny thing is that the current president
of Croatia is a guy named "Stipe Mesic". Not being a great admirer of
Mr. Mesic, I find the thread about his size hilarious. Of course, the
common wisdom is that size doesn't matter, but..., but I've definitely
gone off topic with this.
--
Mladen Gogala
Oracle DBA
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
Other related posts:
