> Akin writes:
> > If I need to setup another network, it definitely will be Linux.
> The proof is certainly in the pudding, eh? I sent Nemo's comments to
> someone I know at a computer magazine for a reaction. If he
> responds, I'll crosspost. I frankly found the 4x test result to be
> quite astounding. Nemo can you tell us about your test?
I did not do the experiment myself. As I mentionned, it was in a
research paper for which I was an anonymous reviewer, therefore I
cannot give the reference. However, I will send you off-line the
e-mail address of one of the authors, and you can ask him for the
paper or permission to reference it -- without saying who you got it
from of course.
In any case, the experiment, conducted between a government
communications research center and a university, was of a
client-server system for delivering video using IP over an ATM
metropolitan area network. They say that with the same hardware, "we
found file tranfer under LINUX to perform roughly 4 times faster than
under DOS/WINDOWS" (exact quote).
> Of course the most valuable comments are from those who have tried
> multiple systems for comparison. We're fortunate at USAID to have a
> shop in which we've got Linux, SCO, NT, DOS, and even Win95 running
> various flavors of servers. I haven't noticed dramatic differences
> in performance across systems. The constraints seem to be more with
> modems and telephone lines these days.
Could be. In the paper I quote, they were able to measure the
difference because, since they were on a highspeed network, the main
bottleneck was at the computers on the two end-points, not in the
network. Of course, if for other reasons (slow modems, or bad phone
lines for example), you cannot transmit data at rates which would
reach the limits of the operating system, then you would not be able
to feel the difference between Windows/NT and Linux.
I think though that, even if they are currently operating with dial-up
connections at 9600 bps all around, by choosing Windoze, service
providers would be self-imposing a speed constraint that they will
regret when they want higher throughput because they have a leased
line going out, or clients connecting over a LAN or MAN, or even a
multiple modem bank.
Another thing is that as Dr. Lisse mentionned, Microsoft does not
follow open standards, so there are other limitations. Recently,
Netscape complained to the government that Windows limited
applications to 10 TCP connections max, so that if you choose to run
server application made by Netscape or someone else, you can have only
ten clients at a time (more if you use Microsoft's own server
application because it uses interfaces to the OS that are not given to
third party developers).