Subject Re: FireBird 1.5 Benchmark
Author Alfred Seetoh
Thanks for your info. Last night I read up about tuning databases,
and I got some insight on the usage of page size. Since then I've
created another database with page size = 8192. Bingo... it works
faster. Thanks again.

--- In firebird-support@yahoogroups.com, Olivier Mascia <om@t...>
wrote:
> Dear,
>
> On Tue, 18 Nov 2003 10:31:38 -0000,
> Alfred Seetoh wrote :
>
> > Forced Writes are enabled for both databases. However I do notice
> > some differences between the 2 of them.
> >
> > Config FB 1.0 FB 1.5
> > ====== ====== ======
> > Page Size 8192 1024
> > ODS Version 10.0 10.1
> > Buffers KB 16384 2048
> >
> > Could the difference page size and buffers cause the difference
in
> > speed?
>
> Of course.
> Your FB 1.5 uses a smaller page size.
> That can impact performance, depending on the mix of operations,
size
> of records, index keys, and so on. Also Win32 OSes like WinNT,
Win2K,
> WinXP on 32 bits computers have a virtual memory page size of 4 KB.
As
> at the lowest level in the OS all disk IO is based on the virtual
> memory subsystem, the natural unit of IO (at the OS level - not
> hardware) is 4 KB. This is why I never create any FB Database on
Win32
> with a page size smaller than 4 KB. Using a NTFS file system with 4
KB
> clusters can be valuable too. But I'm getting off topic now.
>
> Recreate your FB 1.5 database, using a 8 Kb page size, to be fair.
> And test again.
>
> Your number of page buffers is set to 2048 pages in both cases.
> Which gives you 16384 KB for the 8 KB page size, and 2048 KB for
the 1
> KB page size. Nothing abnormal here.
>
> --
> Best Regards,
> Olivier Mascia
> http://www.ibpp.org