Date: Thu, 1 May 1997 15:39:29 +0200 (CEST) From: andreas@klemm.gtn.com (Andreas Klemm) To: WH@ODS.de (Walter Haslbeck), smp@FreeBSD.org Cc: everybodyunix@wup.de, jk@ct.heise.de Subject: Re: Dual PPro Mainboard f. SCO SMP Message-ID: <199705011339.PAA02810@klemm.gtn.com> References: <6VA7t7pLSfB@surfer.ods.de> <6VbVZdMLSfB@surfer.ods.de> <5k1utd$dj8$1@hrz-ws11.hrz.uni-kassel.de> <6VndaKB5SfB@surfer.ods.de>
next in thread | previous in thread | raw e-mail | index | archive | help
[Posted and mailed]
In article <6VndaKB5SfB@surfer.ods.de>,
WH@ODS.de (Walter Haslbeck) writes:
[snip]
> BTW: Wie siehts eigentlich z.Z. an der FreeUnix Front aus in Sachen SMP?
> Gibts irgendwo im Web Benchmarkergebnisse, die SCO SMP mit Linux/SMP
> bzw. FreeBSD/SMP vergleichen?
I'm writing this in english, so I can write this also to the
FreeBSD SMP mailinglist. I suppose the people there are interested
in numbers and figures as well ;-)
I'll put together some nice gif files later, that are suitable to
be put onto a Web Server.
I made something like an application level benchmark by compiling
my FreeBSD custom kernel with different kernels and job parameters
for make (-j). You can see this kind of tests already in PC magazines
like C't, where they bench the compile time for a Linux kernel.
For reference:
Tyan Titan Pro ATX, 64 MB RAM, 2 x 200 MHz Pro
>>> normal -current kernel
make -j 1 262.09 real 189.47 user 15.21 sys
make -j 2 215.58 real 191.09 user 16.24 sys
make -j 4 213.17 real 191.39 user 19.49 sys
make -j 8 213.24 real 191.91 user 19.40 sys
make -j 16 215.19 real 192.10 user 19.31 sys
>>> -current SMP kernel 1 CPU
make -j 1 269.95 real 189.22 user 23.39 sys
make -j 2 222.68 real 191.38 user 23.03 sys
make -j 4 218.24 real 192.00 user 23.47 sys
make -j 8 217.93 real 191.35 user 24.29 sys
make -j 16 220.15 real 191.58 user 24.26 sys
>>> SMP kernel 2 CPUs (after sysctl -w kern.smp_active=2)
make -j 1 252.35 real 144.69 user 75.06 sys
make -j 2 137.06 real 169.23 user 56.46 sys
make -j 4 119.95 real 176.02 user 53.25 sys
make -j 8 119.08 real 175.47 user 54.97 sys
make -j 16 120.03 real 178.42 user 53.32 sys
observations:
1) comparing FreeBSD Uniprocessor vs. SMP kernel (both with 1 CPU)
Using the SMP kernel the system spent a little more time
in kernel mode.
2) Using the SMP kernel with 2 CPU's and make -j 4 and -j 8 is
really fast
The performance boost you get now with FreeBSD SMP
_for_this_kind_of_application is factor 1.9 !!!
To sum up: you nearly double the performance of your system !!!
I think important here is, that every PPro CPU has it's own 2nd
level cache.
Andreas ///
--
powered by
Symmetric MultiProcessor FreeBSD
http://www.freebsd.org/~fsmp/SMP/SMP.html
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199705011339.PAA02810>