Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 25 Jan 2012 23:14:53 +0100
From:      Florian Smeets <flo@FreeBSD.org>
To:        Andriy Gapon <avg@FreeBSD.org>
Cc:        svn-src-head@FreeBSD.org, svn-src-all@FreeBSD.org, src-committers@FreeBSD.org
Subject:   Re: svn commit: r228424 - in head/sys: kern sys
Message-ID:  <4F207EDD.1000407@FreeBSD.org>
In-Reply-To: <4F207B55.4020500@FreeBSD.org>
References:  <201112112102.pBBL21kB068967@svn.freebsd.org> <4F1DE4FF.3080606@FreeBSD.org> <4F207B55.4020500@FreeBSD.org>

next in thread | previous in thread | raw e-mail | index | archive | help
This is an OpenPGP/MIME signed message (RFC 2440 and 3156)
--------------enigF63A6A7E0C371786D9BE842D
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

On 25.01.12 22:59, Andriy Gapon wrote:
> on 24/01/2012 00:53 Florian Smeets said the following:
>> On 11.12.11 22:02, Andriy Gapon wrote:
>>> Author: avg Date: Sun Dec 11 21:02:01 2011 New Revision: 228424 URL:
>>> http://svn.freebsd.org/changeset/base/228424
>>>
>>> Log: panic: add a switch and infrastructure for stopping other CPUs i=
n
>>> SMP case
>>>
>=20
>> Hi,
>=20
>> Attilio asked me to verify that this commit does not introduce a=20
>> performance regression.
>=20
> First of all, thank you very much for doing this!
>=20
>> The box used to run these tests was a 40 Core 32GB Xeon box (HTT was t=
urned
>> off, so 40 real hardware cores). As benchmark pgbench/PostgreSQL were =
used,
>> a snapshot of PostgreSQL 9.2 from 16.01.2012 was used as they did a lo=
t of
>> scaling work in 9.2 which improved the numbers quite a lot vs. 9.1. Th=
e
>> initial benchmarks were run with a scaling factor of 100 which creates=
 a
>> database work set of ~1.5GB. Max throughput was achieved at 20 Clients=
=2E
>=20

[mangled ministat removed]

>=20
>> At 40 threads the results varied between 43000 - 76500 across reboots.=
=20
>> Attilio suspects that this can be caused by the kernel memory layout=20
>> changing under the woods creating cache effects difficult to control, =

>> therefor the scaling factor was reduced to 10 (~150MB work set) and th=
e=20
>> numbers got deterministic across reboot.
>=20

[mangled ministat removed]

>=20
>=20
> Is it possible to see ministat's report for the difference between
> 228424-40-sf10 and 228424-40-sf10-cl datasets?
>=20

Sure, here we go.

x 228424-40-sf10
+ 228424-40-sf10-cl
+----------------------------------------------------------------------+
|x  + +  +     x      ++x   + x +   +      ++  x   x                  x|
|        |____________________M_A_____________________|                |
|         |______________A__M__________|                               |
+----------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stdd=
ev
x  10     194093.35     198864.83     196129.36     196214.69     1545.87=
83
+  10     194288.28     197083.85     195955.26     195733.15     1012.35=
29
No difference proven at 95.0% confidence

>> The 228424-40-sf10-cl results are with a patch from Attilio [1] which =
he=20
>> will followup on.
>=20
> I like the patch already, regardless of how much performance difference=
 it
> produces :-)
>=20

Florian

[1] http://people.freebsd.org/~attilio/cachelineunshare.patch


--------------enigF63A6A7E0C371786D9BE842D
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: OpenPGP digital signature
Content-Disposition: attachment; filename="signature.asc"

-----BEGIN PGP SIGNATURE-----

iEYEARECAAYFAk8gft4ACgkQapo8P8lCvwnSvACghSPbOTJMy4j/4pCGtP/+ZXn8
oOkAn3GQd9LA3SznCIMd1IE57PuQvwtv
=hrC0
-----END PGP SIGNATURE-----

--------------enigF63A6A7E0C371786D9BE842D--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4F207EDD.1000407>