Skip site navigation (1)Skip section navigation (2)
Date:      20 Feb 2001 21:59:46 +0000
From:      John Fremlin <chief@bandits.org>
To:        <freebsd-emulation@FreeBSD.ORG>
Subject:   Re: FreeBSD 4.2 RELEASE twice as slow as Linux
Message-ID:  <m2y9v1t3ql.fsf@boreas.yi.org.>
In-Reply-To: Kris Kennaway's message of "Mon, 19 Feb 2001 16:07:06 -0800"
References:  <"m28zn2kysa.fsf"@boreas.yi.org> <20010219160706.B55565@mollari.cthul.hu>

Next in thread | Previous in thread | Raw E-Mail | Index | Archive | Help

Sorry for all the people I offended with my subject line. I was
wrong. I mistakenly compiled the kernel with a different .config under
FreeBSD. I repeated the test again with the right .config and got only
about a one minute difference. Why so vague you ask? Well, FreeBSD
decided to eat my timings file (which I had foolishly left on my ext2
fs), presumably in retaliation for linux's previously munging my ufs
parition. :-)

Even more annoyingly I also lost the vmstats, so I can only give a
vague idea of my software setup. The only other processes in FreeBSD
were IIRC sshd, usbd, moused(?), syslogd(?) and gettys. In Linux I had
syslogd, klogd, named*5 (BIND 9.1.0), gpm, pppd and gogettys (extra
small getty I wrote). I compiled my FreeBSD kernel with -O2 -march=k6,
to try and balance it out against the Linux 2.4.1 kernel (with some
minor unrelated patches of my own), which I illicitly built with -O2
-fomit-frame-pointer -pipe -fno-strict-aliasing
-mpreferred-stack-boundary=2 -march=k6 -fschedule-insns2
-funroll-loops for luck.

I have a stepping 0 AMD K6-2 300 MHz with one stick of 64 Mb of 100
MHz SDRAM (don't know timings). I have a Alladin 5 IDE
controller. Linux and FreeBSD use the same 128 Mb swap partition on an
old SAMSUNG VG36483A (6.48GB) (ad0). The build was done on an
IBM-DTLA-305020 20 Gb (ad2).

Not so anxious to repeat that particular test now ;-), unless someone
figures they know how to snatch up the 10% (IIRC there was a lot more
systime 1 minute+ against 40 seconds on Linux).

Kris Kennaway <kris@obsecurity.org> writes:

> No - probably there were differences in how you compiled them, what
> compiler you used, what filesystem mount options you used, etc.

The build partition was an ext2fs one:

Under FreeBSD 4.2 RELEASE

/dev/ad2s7	/home/john	ext2fs	rw,noatime	0   	2

And under Linux 2.4.1 glibc-cvs

/dev/hdc7	/home/john	ext2	defaults,noatime	0   	2

The build environment was linux ELF native on ext2fs in both cases

/dev/ad2s5	/mnt/root	ext2fs	ro,nodev,noatime	0   1

and

/dev/hdc5	/	ext2	defaults,noatime	0   1

> Such a naieve benchmark doesn't really show anything.

What benchmark do you think would show FreeBSD (linux emulation) in
better light? I only started on this because someone developing the
linux kernel VM said that memory management made little difference to
linux kernel compile times, and I wanted to show the contrary by using
FreeBSD as a counterexample (honest).

-- 

	http://www.penguinpowered.com/~vii

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-emulation" in the body of the message




Want to link to this message? Use this URL: <http://docs.FreeBSD.org/cgi/mid.cgi?m2y9v1t3ql.fsf>