Message-ID: <457FC628.50207@aeccom.com>
Date: Wed, 13 Dec 2006 10:21:44 +0100
From: Sven Geisler <sgeisler@aeccom.com>
Organization: AEC/communications GmbH
User-Agent: Thunderbird 1.5.0.8 (Windows/20061025)
MIME-Version: 1.0
To: Daniel van Ham Colchete <daniel.colchete@gmail.com>
CC: pgsql-performance@postgresql.org
Subject: Re: New to PostgreSQL, performance considerations
References: <60psaq5t94.fsf@dba2.int.libertyrms.com>	
	<E1Gtplp-0004Ka-GM@elasmtp-junco.atl.sa.earthlink.net>	
	<20061211184726.GI16692@mathom.us>	
	<E1GtrB8-0005rw-Iu@elasmtp-kukur.atl.sa.earthlink.net>	
	<457DA782.3060506@modgraph-usa.com>	
	<8a0c7af10612111422x6bdcc0c2p7cbfeb5a7b0e0374@mail.gmail.com>	
	<20061212013706.GM16692@mathom.us>	
	<8a0c7af10612120110v34f7d3c3p6d872efabf97b922@mail.gmail.com>	
	<20061212123131.GQ16692@mathom.us>	
	<8a0c7af10612120858g1b1b4454x518a3ed609324627@mail.gmail.com>
	<8a0c7af10612121035g26a00862qcc34284d76d5230a@mail.gmail.com>
In-Reply-To: <8a0c7af10612121035g26a00862qcc34284d76d5230a@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit

Hi,

Did someone try '-mfpmath=sse -msse3'?

Would be interesting to know if -mfpmath=sse boost the performance.

I guess, the difference in the generated code isn't that much between
i686 and prescott. The bigger step is i386 to i686. '-mfpmath=sse
-msse3' will also use the SSE unit, which the classic i686 doesn't have.

CFLAGS=-O2 -march=prescott -mfpmath=sse -msse3

Best regards
Sven.

Daniel van Ham Colchete schrieb:
> I just made another test with a second Gentoo machine:
> 
> Pentium 4 3.0Ghz Prescott
> GCC 4.1.1
> Glibc 2.4
> PostgreSQL 8.1.5
> Kernel 2.6.17
> 
> Same postgresql.conf as yesterday's.
> 
> First test
> ==========
>  GLIBC: -O2 -march=i686
>  PostgreSQL: -O2 -march=i686
>  Results: 974.638731 975.602142 975.882051 969.142503 992.914167
> 983.467131 983.231575 994.901330 970.375221 978.377467
>  Average (error):  980 tps (13 tps)
> 
> Second test
> ===========
>  GLIBC: -O2 -march=i686
>  PostgreSQL: -O2 -march=prescott
>  Results: 988.319643 976.152973 1006.482553 992.431322 983.090838
> 992.674065 989.216746 990.897615 987.129802 975.907955
>  Average (error):  988 tps (15 tps)
> 
> Third test
> ==========
>  GLIBC: -O2 -march=prescott
>  PostgreSQL: -O2 -march=i686
>  Results: 969.085400 966.187309 994.882325 968.715150 956.766771
> 970.151542 960.090571 967.680628 986.568462 991.756520
>  Average (error): 973 tps (19 tps)
> 
> Forth test
> ==========
>  GLIBC: -O2 -march=prescott
>  PostgreSQL: -O2 -march=prescott
>  Results: 980.888371 978.128269 969.344669 978.021509 979.256603
> 993.236457 984.078399 981.654834 976.295925 969.796277
>  Average (error):  979 tps (11 tps)
> 
> The results showed no significant change. The conclusion of today's
> test would be that there are no improvement at PostgreSQL when using
> -march=prescott.
> 
> I only see 3 diferences between yesterday's server and today's: the
> kernel version (y: 2.6.18, t:2.6.17), the server uses an IDE harddrive
> (yesterday was SATA), and the gcc version (3.4.6 -> 4.1.1).
> 
> I don't know why yesterday we had improved and today we had not.
> 
> Best
> Daniel
> 
> On 12/12/06, Daniel van Ham Colchete <daniel.colchete@gmail.com> wrote:
>> I'm making some other tests here at another hardware (also Gentoo). I
>> found out that PostgreSQL stops for a while if I change the -t
>> parameter on bgbench from 600 to 1000 and I have ~150 tps instead of
>> ~950tps.