Nico Callewaert wrote:

> Hi again,
>
> I don't know if I have to disagree with Paul, but as a test I backuped and
> restored the database and I changed the buffersize from 4096 to 2048 and it
> looks slower to me now ???
>

Hi, Nico. My English don't allow me to give full and clear explanation. But. Size
of the page mainly influent on performance by 2 aspects
1. How much index entries can be placed on the page (therefore how many pages should
be visited when building access bitmaps). This means large page is better.
2. OS and hardware expensies to read larger number of bytes when need little of page
contens. This means small page is better.

Good compromise to start is set page size equal to HDD cluster size and check
statistics on indexes. If index depth>3 then it is ineffective and experiments should
be continued, try increase page size and check index'es depth and subjective
performance. I have read somewhere that on NTFS page size > 4096 is bad, but can't
remember where and don't tried - I use Linix to run SQL servers.

Sorry if simplified somewhat to point of incorrection but general direction is
certainly right.

Best regards.