Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 7 Feb 2000 23:55:05 +0200
From:      Giorgos Keramidas <charon@hell.gr>
To:        Stanley Hopcroft <Stanley.Hopcroft@IPAustralia.Gov.AU>
Cc:        FreeBSD-Questions@FreeBSD.org, FreeBSD-ISP@FreeBSD.org
Subject:   Re: Performance of FreeBSD and MS Windows. What about select() and memory management etc ?
Message-ID:  <20000207235505.B8424@hades.hell.gr>
In-Reply-To: <Pine.BSF.4.21.0002071050190.6891-100000@stan>; from Stanley.Hopcroft@IPAustralia.Gov.AU on Mon, Feb 07, 2000 at 11:40:43AM %2B1100
References:  <Pine.BSF.4.21.0002071050190.6891-100000@stan>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Feb 07, 2000 at 11:40:43AM +1100, Stanley Hopcroft wrote:
> 
> I am writing to ask about the relative performance of FreeBSD and MS
> Windows 95 and NT as desktop and server.
>
> This letter is inspired by my own experience of FreeBSD 3.4-RELEASE
> and Win 95 OSR2 as a desktop on the same hardware (P5 166MHz, slow
> IDE, 32 MB RAM) and ariticles about Intel Unix (Linux) in magazines.

So, to keep ourselves from making a *huge* mistake here, taking the
observations below as objective, this is about personal experience.
It's a good thing you're using a given hardware setup for making the
comparisons below.  However, the "facts" shown below should be taken as
nothing more than they really are, that is... your own experience.

Thank you Stanley for clarifying this in the VERY first place, before
making *any* observation at all.

> My experience with FreeBSD 3.4-RELEASE (kde 1.1.2, Communicator 4.7,
> Metaframe, pine etc) as a desktop is that it seems to thrash more
> than OSR2. I think this is because Netscape wants to have 20 or MB of
> memory.

I would certainly love to hear more about this thrashing you're
referring to.  It might be because of my lack of luck, or because of a
weird combination of circumstances, but I haven't actually managed to
make my FreeBSD workstation thrash, and my setup is similar to yours,
alas even worse (P5 133 MHz, 32 Mb EDO RAM).

> In "Windows NT magazine" (May 1999), an article "Linux and the
> Enterprise: is this OS ready for prime time" by Mark Russinovich,
> compares the network performance of NT and Linux (or other Posix 1
> compliant OS) unfavourably on the basis that select() does not scale,
> or perform as well as the non- standard system call that MS provides
> and the author claims is implemented on other high performance Unix
> platforms.

Is this article available on the Web anywhere?  One thing is that, I
would be interested to see the actual data of this comparison.  The
other, probably more important, thing is that this article is comparing
_Linux_ to Windows NT.

It might seem a bit irrelevant, but FreeBSD has nothing to do with
Linux, neither in terms of source, nor in terms of stability,
robustness, etc.  I have yet to see a Linux server that can handle
a lot more than 500-1000 simultaneous connections; but there a few
FreeBSd installations that can serve several thousand of connections
simultaneously, with ftp.cdrom.com and www.yahoo.com being among the
most prominent ones.

What is more interesting in that article is that it seems to favour the
"non-standard" Microsoft system calls, which are, well, non-standard.
There is nothing wrong with making incompatible changes to something
that has been considered "standard" for a long time, and the BSD-socket
system calls are one of the aged standards we all know nowadays.
However, if you want everyone to adopt and benefit from revolutionary
ideas that break existing standards, it's better if you let those ideas
open to everyone.  I might be utterly wrong here, but I don't seem to
have heard of any open extensions to BSD sockets announced by anyone,
especially not Microsoft.

> The same author in another article claims that the VM system of
> Unix also fails to provide the facilities or performance of the MS
> system. He claims his conclusions are based on his inspection of the
> Linux kernel, and I presume, what MS claim about their kernel.

Generalizations like using the word "Unix" to talk about a whole family
of operating systems are nice, but they can also result in misleading
statements and conclusions.

As far as I know the VM subsystem of each operating system is quite
different from one operating system to the other.  Sometimes, even the
same operating system has major changes on it's VM management among
different versions.

Now, the VM system of Linux is not "the best around the block", but it
has one major advantage over the claims of anyone.  The source that
implements it is open to review by anyone with adequate knowledge of
programming in C and a working knowledge of x86 assembler.  I tend to
believe more easily someone whose source code I can easily access, than
anyone whose sentences start with "trust me".

> As for me, I will show more interest in the incredibly high performace
> and sophistication of MS Windows when the products are more usable (as
> servers) and available - in any thing other than a file server role.

On the machine setup that you have described, I would be tremendously
indebted to know what one can do to achieve this incredibly high
performance.  Because, contrary to my own openness to hear other
people's opinions, my own experience seems to indicate the opposite.

I am always interested in hearing about facts accompanied with the
appropriate amount of data.  For instance, saying that Netscape takes a
bit more to load in FreeBSD than Windows, might be a nice thing to know.
A nicer thing to know however is the time that subsequent instances of
Netscape take to load, or the memory footpring of each new instance.

Oh, and that would probably mean nothing as far as performance is
concerned, because Netscape is the last thing I would start running on a
file server.  I can think of better uses for the disk space of a
file-server, quite better than storing Netscape's huge Motif-dependent
binaries.

-- 
Giorgos Keramidas, < keramida @ ceid . upatras . gr >
For my public PGP key: finger keramida@diogenis.ceid.upatras.gr
PGP fingerprint, phone and address in the headers of this message.


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




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000207235505.B8424>