*BSD News Article 65256


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!newshost.telstra.net!act.news.telstra.net!psgrain!usenet.eel.ufl.edu!spool.mu.edu!howland.reston.ans.net!newsfeed.internetmci.com!news.sprintlink.net!new-news.sprintlink.net!news.clark.net!news.clark.net!not-for-mail
From: gsh@clark.net (Greg Hennessy)
Newsgroups: comp.os.linux.development.apps,comp.os.linux.development.system,comp.os.linux.x,comp.os.linux.hardware,comp.os.linux.setup,comp.unix.bsd.386bsd.misc,comp.unix.bsd.bsdi.misc,comp.unix.bsd.netbsd.misc,comp.unix.bsd.freebsd.misc
Subject: Re: Why to not buy Matrox Millennium
Date: 5 Apr 1996 14:15:54 -0500
Organization: Clark Internet Services, Inc., Ellicott City, MD USA
Lines: 48
Message-ID: <4k3rha$m9t@clark.net>
References: <4jn4qp$6p@darkstar.my.lan> <4jve3t$cfe@hermes.synopsys.com> <4k0m0f$68j@hoopoe.psc.edu> <4k0r5l$g2@siberia.gtri.gatech.edu>
NNTP-Posting-Host: explorer.clark.net
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=ISO-8859-1
Content-Transfer-Encoding: 8bit
Xref: euryale.cc.adfa.oz.au comp.os.linux.development.apps:14232 comp.os.linux.development.system:20815 comp.os.linux.x:28755 comp.os.linux.hardware:35639 comp.os.linux.setup:49301 comp.unix.bsd.386bsd.misc:487 comp.unix.bsd.bsdi.misc:3038 comp.unix.bsd.netbsd.misc:2805 comp.unix.bsd.freebsd.misc:16787

>I have a couple of other questions, in addition to whether these results
>were obtained on platforms of comparable "general" performance: basically
>to know whether the performance was attributable to the system architecture
>(i.e., did the processor have to swap bytes and restructure the packets?),
>or the os (were there other x86 comparisons than FreeBSD vs. Linux).

The paper benchmarked Solaris 2.4, FreeBsd 2.0.5R, and Linux 1.2.8. 

All were installed on the same 100Mhz Pentium with 32 meg ram, and two
2 gig disks, and a 3Com 3c509 ethernet card. 

The url of the paper is
http://mosquitonet.stanford.edu/~laik/benchmarks/index.html. 

>Especially since the early tenor of this thread seemed to
>suggest that Linux basic quality was at the "toy" level.

Someone was engaged in a bit of Linux bashing, yes. 

>I'm hoping that more information will identify _why_ this results.

The paper shows that Liux has a maximum UDP bandwidth of ~16 megbits per
second. FreeBsd acheives 50 megbits/sec, while Solaris acheived
32megbits/sec. 

Note these are values from talking to the loopback device, NOT an
acutal ethernet card.

Under TCP Freebsd had 65 megbits/sec, Solaris 60megbits/sec, and Linux
25 megbits per sec. 

The percentage values that the earlier poster quoted are because Linux
gets a much higher bandwith in a pipe. If you divide the actual
throughput by the pipebandwidth, linxus fares worse than the other
systems. But that is dividing an apple by an orange and comparing the
result to a bananna. Quoting 14% udp bandwidth for linux is linux
bashing off an inappropiate metric. 

The two regions where Linux came out behind FreeBSD are in the
scheduler and in networking. Both of these areas are much improved in
the linux 1.3.* kernels.

The areas where FreeBSD came out behind Linux are in disk IO.
Apparently the newest version of FreeBSD will have improvements there.


An interesting article.