Daniel van Ham Colchete wrote:
> On 12/11/06, Steinar H. Gunderson <sgunderson@bigfoot.com> wrote:
>> On Mon, Dec 11, 2006 at 11:31:48AM -0200, Daniel van Ham Colchete wrote:
>> > What PostgreSQL benchmark software should I use???
>>
>> Look up the list archives; search for "TPC".
>>
>> > I'll test PostgreSQL 8.1 on a Fedora Core 6 and on a Gentoo. I'll get
>> > the same version FC6 uses and install it at my Gentoo. I'll use the
>> > same hardware (diferent partitions to each).
>>
>> Why do you want to compare FC6 and Gentoo? Wasn't your point that the
>> -march=
>> was supposed to be the relevant factor here? In that case, you want to
>> keep
>> all other things equal; so use the same distribution, only with -O2
>> -march=i686 vs. -march=athlon-xp (or whatever).
>>
>> /* Steinar */
> Using Gentoo is just a easy way to make cflag optimizations to all the
> other libs as well: glibc, ...
>
> I can also mesure performance on Gentoo with cflag optimized
> PostgreSQL and plain PostgreSQL as well.
>
I can certainly recall that when I switched from Fedora Core (2 or 3
can't recall now) to Gentoo that the machine was "faster" for many
activities. Of course I can't recall precisely what now :-(.
To actually track down *why* and *what* make it faster is another story,
and custom CFLAGS is only 1 of the possible factors: others could be:
- different kernel versions (Gentoo would have possibly been later)
- different kernel patches (both RedHat and Gentoo patch 'em)
- different versions of glibc (Gentoo possibly later again).
- different config options for glibc (not sure if they in fact are, but
it's possible...)
- kernel and glibc built with different versions of gcc (again I suspect
Gentoo may have used a later version)
So there are a lot of variables to consider if you want to settle this
debate once and for all :-)!
Best wishes
Mark
