DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.co.uk;
	h=Message-ID:X-YMail-OSG:Received:X-Mailer:Date:From:Subject:To:Cc:In-Reply-To:MIME-Version:Content-Type:Content-Transfer-Encoding;
	b=vIhQYXPYP0ljVWPScJqLe+uCtUjpin5KpalLZYVBMFrWi9nzIPO0PiaAS2d//QBtwqN317T3j4m+GfjyYOETgdIEsL0GwH0wYvPskIDaNvX3dT5mfKEXf7G64CLc304ABT90NdDeQzg2iGFXt7H9+qK2zWeSFbpnFhW0spAfMVU=;
Message-ID: <527122.4074.qm@web26005.mail.ukl.yahoo.com>
Date: Tue, 12 Apr 2011 13:35:19 +0100 (BST)
From: Glyn Astill <glynastill@yahoo.co.uk>
Subject: Re: Linux: more cores = less concurrency.
To: david@lang.hm, Steve Clark <sclark@netwolves.com>,
	Kevin Grittner <Kevin.Grittner@wicourts.gov>
Cc: "Joshua D. Drake" <jd@commandprompt.com>,
	Scott Marlowe <scott.marlowe@gmail.com>, pgsql-performance@postgresql.org
In-Reply-To: <4DA32DE3020000250003C68E@gw.wicourts.gov>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: quoted-printable

--- On Mon, 11/4/11, Kevin Grittner <Kevin.Grittner@wicourts.gov> wrote:=0A=
=0A> From: Kevin Grittner <Kevin.Grittner@wicourts.gov>=0A> Subject: Re: [P=
ERFORM] Linux: more cores =3D less concurrency.=0A> To: david@lang.hm, "Ste=
ve Clark" <sclark@netwolves.com>, "Kevin Grittner" <Kevin.Grittner@wicourts=
.gov>, "Glyn Astill" <glynastill@yahoo.co.uk>=0A> Cc: "Joshua D. Drake" <jd=
@commandprompt.com>, "Scott Marlowe" <scott.marlowe@gmail.com>, pgsql-perfo=
rmance@postgresql.org=0A> Date: Monday, 11 April, 2011, 22:35=0A> "Kevin Gr=
ittner" <Kevin.Grittner@wicourts.gov>=0A> wrote:=0A>  =0A> > I don't know w=
hy you were hitting the knee sooner than=0A> I've seen=0A> > in my benchmar=
ks=0A>  =0A> If you're compiling your own executable, you might try=0A> boo=
sting=0A> LOG2_NUM_LOCK_PARTITIONS (defined in lwlocks.h) to 5 or=0A> 6.=A0=
 The=0A> current value of 4 means that there are 16 partitions to=0A> sprea=
d=0A> contention for the lightweight locks which protect the=0A> heavyweigh=
t=0A> locking, and this corresponds to your best throughput=0A> point.=A0 I=
t=0A> might be instructive to see what happens when you tweak the=0A> numbe=
r=0A> of partitions.=0A>  =0A=0ATried tweeking LOG2_NUM_LOCK_PARTITIONS bet=
ween 5 and 7. My results took a dive when I changed to 32 partitions, and i=
mproved as I increaced to 128, but appeared to be happiest at the default o=
f 16.=0A=0A> Also, if you can profile PostgreSQL at the sweet spot and=0A> =
again at a=0A> pessimal load, comparing the profiles should give good=0A> c=
lues about=0A> the points of contention.=0A>  =0A=0AResults for the same ma=
chine on 8 and 32 cores are here:=0A=0Ahttp://www.8kb.co.uk/server_benchmar=
ks/dblt_results.csv=0A=0AHere's the sweet spot for 32 cores, and the 8 core=
 equivalent:=0A=0Ahttp://www.8kb.co.uk/server_benchmarks/iostat-32cores_32C=
lients.txt=0Ahttp://www.8kb.co.uk/server_benchmarks/vmstat-32cores_32Client=
s.txt=0A=0Ahttp://www.8kb.co.uk/server_benchmarks/iostat-8cores_32Clients.t=
xt=0Ahttp://www.8kb.co.uk/server_benchmarks/vmstat-8cores_32Clients.txt=0A=
=0A... and at the pessimal load for 32 cores, and the 8 core equivalent:=0A=
=0Ahttp://www.8kb.co.uk/server_benchmarks/iostat-32cores_100Clients.txt=0Ah=
ttp://www.8kb.co.uk/server_benchmarks/vmstat-32cores_100Clients.txt=0A=0Aht=
tp://www.8kb.co.uk/server_benchmarks/iostat-8cores_100Clients.txt=0Ahttp://=
www.8kb.co.uk/server_benchmarks/vmstat-8cores_100Clients.txt=0A       =0Avm=
stat shows double the context switches on 32 cores, could this be a factor?=
 Is there anything else I'm missing there?=0A=0ACheers=0AGlyn