Re: 8k block size upgrade concern

From: Fabrizio <fabrizio.magni_at_mycontinent.com>
Date: Wed, 29 Jun 2005 10:26:08 +0200
Message-ID: <42c25aa8$0$62494$892e7fe2@authen.white.readfreenews.net>

DA Morgan wrote:
> JeffV wrote:
>

>> Daniel,
>>
>>
>> Are you suggesting then that there should be a benefit to upgrading
>> Oracle 4k to 8k since 8k matches the operating system block size?
>>
>> Thanks,
>>
>> Jeff

>
>
> The default block size in Oracle should ALWAYS be the block size of
> the operating system, as installed, unless actual testing with your
> application and your data demonstrates otherwise.
>
> The advice that has been given in a few books, a few magazines, and
> a lot of web postings about smaller blocks sizes for OLTP and larger
> for OLAP is not demonstrated by testing and should be ignored.

Sorry to intrude but I find the post a little bit misleading.

The rule of thumb you are referring (the one coming from Steve Adams, I believe) is to match the oracle block size to the filesystem block size.

The latter can be different from the operating system page size (sometime referred as OS block size).
In fact the fs block size can be (generally on all unixes) equal or smaller than pagesize (so, for example, you can have a 2k blocksize on solaris).

Always following the same rule of thumb if you are on raw devices (which probably the OP is not), or using direct i/o, you are not constrained for the blocksize choice.

Sincerelly I believe that readahead should be considered more in this kind of choice... just a personal opinion but I'll try to formalize it on august with a paper and benchmarks supporting the thesis... or if I'm wrong with a retraction and an apologize.

Regards

-- 
Fabrizio Magni

fabrizio.magni_at_mycontinent.com

replace mycontinent with europe

Received on Wed Jun 29 2005 - 03:26:08 CDT