Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 3 Dec 1996 08:33:48 -0500 (EST)
From:      "John S. Dyson" <toor@dyson.iquest.net>
To:        torvalds@cs.Helsinki.FI (Linus Torvalds)
Cc:        jkh@time.cdrom.com, davem@jenolan.rutgers.edu, dyson@FreeBSD.org, dennis@etinc.com, kpneal@pobox.com, hackers@FreeBSD.org, lm@engr.sgi.com, iain@sbs.de, sparclinux@vger.rutgers.edu
Subject:   Re: TCP/IP bandwidth bragging
Message-ID:  <199612031333.IAA01138@dyson.iquest.net>
In-Reply-To: <Pine.LNX.3.95.961203093548.16867D-100000@linux.cs.Helsinki.FI> from "Linus Torvalds" at Dec 3, 96 09:54:49 am

next in thread | previous in thread | raw e-mail | index | archive | help
>
> PS. Larry is hopefully making a new version of lmbench sometime in the
> future. And guess what? I actually _hope_ that it shows that Linux has
> some problems. I'm not in this game to bash on others, I'm in it to make
> the best damn system there is. 
> 

I use LMBENCH for the same reason on FreeBSD, and it is useful, but I don't
think that it is valuable to use in marketing hype.  Frankly, the discussions
that I had with you caused me to waste time working on low level
perf -- and in reality gained nothing but improved lmbench results.  The
system is not significantly faster in the real world for the reason of that
specific work.  Yet with a different view of things, I have done more for real
world performance in a few days what two months of low level optimizations
did.

I believe that it is best to let the users who care about perf and
stability to test/use FreeBSD and Linux side by side.  We win
alot of those comparisons hands-down.  If people don't care about
performance, then they can choose NT or Linux or whatever their
evangelist coworkers use (to keep peer pressure under control.)
We have some OS/2 people at work, and they are irritating, not because
their OS isn't good, but because they are just irritating trying to
say how good their OS is.

FreeBSD has vulnerabilities also, but citing lmbench results tells such
a small part of the story that it almost misinforms.  Many times a
low level benchmark that shows a 10% difference actually means a .1%
difference in the real world, and other factors overshadow that easily.
I know how to interpret lmbench, but many people out there might think
that it is the be-all end-all of benchmarks.  Horsepower in a car isn't
either.  It is how the car works (at least for me.)

I have heard claims that low level performance measures some kind of
"quality.",  so be it, I measure quality by things (programs, not
necessarily benchmarks) working and working quickly.

John




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199612031333.IAA01138>