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=FgXCfk0ad8oIiJjpGQJzfq78e/KFDnJrtLtBZdS71Nns8JzCBfer5XAL4CgnAZ13hq5YlSO+uh4Kh5b36c2AeHGrnuqZKJRYIlXxCuTK//yqFl4TyrQyg6rt+xAiin215TVcAOrJRyPAEheuoc+Hvj0fx1plRg79mrv/7FW3WX4=;
Message-ID: <364861.84268.qm@web26005.mail.ukl.yahoo.com>
Date: Tue, 12 Apr 2011 09:57:07 +0100 (BST)
From: Glyn Astill <glynastill@yahoo.co.uk>
Subject: Re: Linux: more cores = less concurrency.
To: Scott Marlowe <scott.marlowe@gmail.com>
Cc: pgsql-performance@postgresql.org
In-Reply-To: <BANLkTikj3hMK4uysAU=BFVTps7OwKUwpog@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: quoted-printable

--- On Tue, 12/4/11, Scott Marlowe <scott.marlowe@gmail.com> wrote:=0A=0A> =
From: Scott Marlowe <scott.marlowe@gmail.com>=0A> Subject: Re: [PERFORM] Li=
nux: more cores =3D less concurrency.=0A> To: "Glyn Astill" <glynastill@yah=
oo.co.uk>=0A> Cc: pgsql-performance@postgresql.org=0A> Date: Tuesday, 12 Ap=
ril, 2011, 6:55=0A> On Mon, Apr 11, 2011 at 7:04 AM, Glyn=0A> Astill <glyna=
still@yahoo.co.uk>=0A> wrote:=0A> > Hi Guys,=0A> >=0A> > I'm just doing som=
e tests on a new server running one=0A> of our heavy select functions (the =
select part of a plpgsql=0A> function to allocate seats) concurrently.=A0 W=
e do use=0A> connection pooling and split out some selects to slony=0A> sla=
ves, but the tests here are primeraly to test what an=0A> individual server=
 is capable of.=0A> >=0A> > The new server uses 4 x 8 core Xeon X7550 CPUs =
at=0A> 2Ghz, our current servers are 2 x 4 core Xeon E5320 CPUs at=0A> 2Ghz=
.=0A> >=0A> > What I'm seeing is when the number of clients is=0A> greater =
than the number of cores, the new servers perform=0A> better on fewer cores=
.=0A> =0A> O man, I completely forgot the issue I ran into in my=0A> machin=
es, and=0A> that was that zone_reclaim completely screwed postgresql=0A> an=
d file=0A> system performance.=A0 On machines with more CPU nodes=0A> and h=
igher=0A> internode cost it gets turned on automagically and=0A> destroys=
=0A> performance for machines that use a lot of kernel cache /=0A> shared=
=0A> memory.=0A> =0A> Be sure and use sysctl.conf to turn it off:=0A> =0A> =
vm.zone_reclaim_mode =3D 0=0A> =0A=0AI've made this change, not seen any im=
mediate changes however it's good to know. Thanks Scott.