Re: formulas for correct sizing of various parameters

On Sat, 1 Sep 2001 09:52:12 +1000, "Howard J. Rogers" <howardjr_at_www.com> wrote:

>
>But as Steve also points out, a larger block size makes for huge performance
>improvements on index reads -and what do OLTP systems use to retrieve the
>small bits of data they want to access? Yup -indexes.

and I have yet to find an OLTP system that at some stage doesn't have a large batch component. Even if it's just to collect daily stuff and shove it somewhere else. Some are even designed to be mixed environments. Any of these will benefit more from an 8K block size than they might loose from it.

In UNIX, 8K makes a *lot* of sense from just about any angle. In NT, I have my serious doubts with Steve's 16K recommendation. I've never been able to make it work any better than just setting cluster sizes to 4K or 8K and go with those for db block size as well. Still, quite prepared to admit I'm doing something wrong or measuring results the wrong way.

Cheers
Nuno Souto
nsouto_at_optushome.com.au.nospam Received on Fri Aug 31 2001 - 20:28:04 CDT