Re: os io unit , db block size relation

From: FM <fabrizio.magni_at_mycontinent.com>
Date: Fri, 17 Sep 2004 19:24:05 GMT
Message-ID: <414B39D4.6080806@mycontinent.com>

Yong Huang wrote:

>
> Using iostat (wsec/s and wrqm/s) or (rsec/s and rrqm/s) values to
> calculate driver write or read chunk size is more accurate than what I
> suggested in my last message.
>
> A slightly off-topic question. What if the filesystem block size is
> not 4k while PAGE_SIZE is 4096? (I had this question before for
> Solaris http://groups.google.com/groups?selm=b3cb12d6.0203181445.f3d2b36%40posting.google.com)
>

Not so "off-topic".
The file system block size is normally limited by "page size". So you cannot have bigger blocks than your system pagesize but you can have them smaller.
It will risult in better space allocation but worst performance due to the overhead for the increasing i/o calls.

I'm not sure but if my mind is not falling apart solaris and hp-ux should have a default of 8k. Aix and linux 4k.

You can have bigger pagesize with 64 bit architectures. ia64 on linux is an example (default 16k).

>

>>In tolga's example he wasn't writing on a real disk so the metrics 
>>weren't really valid.

>
>
> You mean he's writing to RAID? Shouldn't RAID be transparent to
> iostat?
>
>

No, it wasn't my point. In his test he was writing on the pseudo device /dev/null.

>
> Thanks for the compliments. Some of the stuff on my site is meant to
> attract recruiters. Real gurus may laugh at it.
>

Your site attracted me (it was one of the first I found, a year and a half ago, when I started learning oracle) and you deserve praise for that job.

-- 

Fabrizio Magni

fabrizio.magni_at_mycontinent.com

replace mycontinent with europe

Received on Fri Sep 17 2004 - 14:24:05 CDT