Re: Performance difference between 2 machines

From: DA Morgan <damorgan_at_psoug.org>
Date: Tue, 21 Jun 2005 11:00:46 -0700
Message-ID: <1119376851.311996@yasure>

Matthias Hoys wrote:
> "DA Morgan" <damorgan_at_psoug.org> wrote in message
> news:1119307340.321454_at_yasure...
>

>>Matthias Hoys wrote:
>>
>>
>>>But, if I use a (low) block size of 4k, won't I risk ending up with lots 
>>>of chained rows for "wide" tables (and thus double i/o to get those rows) 
>>>?
>>
>>Possible. But we have no way of knowing what your average and maximum
>>row lengths are from what you have posted. Or if your tables are
>>properly normalized or have been flattened and thus have row sizes far
>>larger than they should be.
>>
>>You need to work these things out for yourself based on analysis of your
>>application.
>>-- 
>>Daniel A. Morgan
>>http://www.psoug.org
>>damorgan_at_x.washington.edu
>>(replace x with u to respond)

>
>
> Regarding the AIX filesystem block size : when I look at the properties of
> our enhanced journaled filesystems (JFS2), the unit size = 512 bytes and the
> filesystem block size = 4096 bytes. So what's the difference between unit
> size and block size ?
> One of our 8.1.7 DWH databases (which needs to be migrated to Oracle 10g)
> has a block size of 16k, so what would be the reason for this ? (it was
> installed a long time ago by an external consultant).
> Is there any way to estimate the optimal block size for an existing database
> ?
>
>
> Matthias

The reason for the 16K block size? Perhaps someone has been reading too many books by authors that haven't backed up their theoretical projections with actual benchmarking.

You should be using 4K blocks with AIX.

-- 
Daniel A. Morgan
http://www.psoug.org
damorgan_at_x.washington.edu
(replace x with u to respond)

Received on Tue Jun 21 2005 - 13:00:46 CDT