Re: OS block size

"Dave Haas" <davidh_at_--nospam--hotmail.com> wrote in message news:cW9w6.67821$tr5.7039641_at_news1.telusplanet.net...
> Hi Howard.
>
> > Utter nonsense (and we've been round this one time and time again). See
> > Steve Adams' web site for his article 'Why a Large Database Block Size':
> > Most Unixes should use 8K blocks, and NT should be using 16K or so.
 Block
> > size is definitely an O/S issue, not a 'what do I want to use this
 database
> > for' issue.
>
> I disagree. Using 16k blocks in an OLTP environment on an NT system would
> quite probably introduce contention issues.

Undoubtedly. But they can be managed, as I said.

>
> >
> > Block contention is most definitely a worry for OLTP systems. Curing it
 by
> > adopting a small block size is rather like curing a headache by cutting
 off
> > your head. There are other, better and more scientific methods of
 managing
> > the issue (initrans and maxtrans springs to mind).
>
> That's a little harsh. There are many methods to preventing and curing
> contention and I would argue that choice of block size is definately one
 of
> them.
>

It's not a valid (let's rather say a 'reasonable') choice, however, since by doing so you set yourself up to fight the natural instincts of the file system on which you are running.

However, this is an extremely old (and tired) argument, I'm out of here shortly, and although it saddens me that these old wive's tales are still doing the rounds, I've gotten to the point where, like Rhett, I couldn't give a damn!

Do, please, read Steve's article on the subject, for it is cogent and unambiguous, and well-argued:
http://www.ixora.com.au/tips/creation/block_size.htm

Regards
HJR
> Regards,
>
> Dave
>
>
>
Received on Wed Mar 28 2001 - 02:15:17 CST