[ale] News from the front: The O/S Wars Continue

[JH]

It becomes rather ridiculous after a while, doesn't it?  The fact remains that
you can use Linux/Samba to make a FAST Web or Samba server.  And where did they
get the bit about 4 NICs??  What I think is cool is that you can make a pretty
fast Samba server out of "junk" with Linux, provided you've set things up such
that most of the time, the same files are being read out of Samba shares by SMB
clients over and over again, thereby taking advantage of Linux' ability to
manage its disk cache quite nicely.  This is the sort of thing that would happen
if you were using a Linux/Samba server to hold Windows apps.  I messed with this
myself using an Overdrive/100 PCI system with an ancient 1.6GB Micropolis SCSI
drive.  The drive was so slow that the first time you tried to launch a Windows
app from it, you could feel the lag...but the second time, ZAP!  No faster
machine would have really made any difference to anybody (my term for that is
"arbitrarily fast").  If you had an environment with a few hundred clients, you
could probably arrange things so that most of the time, everybody got that kind
of performance, even if the server hardware isn't exactly cutting-edge.

Legacy hardware re-use is one of the things I am keenly interested in, and Linux
helps make it possible.

- Jeff

Chris Ricker wrote:

> On Mon, 28 Jun 1999, Ryan Bridges wrote:
>
> > Read this article.  It is not a rigged test but a true benchmark with
> > arguments supporting their benchmarks.  Everyone should really read this.
>
> I disagree that it's a "true benchmark."  It suffers from the same problem
> the first one ultimately did (modulo mindcraft's incompetence in tuning
> Linux ;-), which is that it has absolutely no correspondence to anything
> like reality. No one wastes that kind of bandwidth on small static content.
> No one who can afford the bandwidth Linux would saturate with 4 100Mbit
> cards would use 100Mbit instead of Gigabit.  The benchmark did absolutely
> nothing to simulate real-world 'net traffic patterns (using NIST or such).
> Etc.
>
> It is useful in that it exposed problems that needed to be fixed, and I'm
> glad that ZD has been presenting their writeups in that tone.  At the same
> time, it's utterly worthless as a benchmark.
>
> later,
> chris
>
> --
> Chris Ricker                                               kaboom at gatech.edu
>                                               chris.ricker at genetics.utah.edu