See also: last week's Back page page.

Linux links of the week

Over a year ago, in February of 1998, we mentioned the SANE project as one of our first "Links of the Week". SANE, or Scanner Access Now Easy is still an essential resource for anyone wanting to use a scanner with Linux. They are up to version 1.01 and are due major kudos for the service they've done for the community.

Section Editor: Jon Corbet

Letters to the editor

From: Brett Viren <bviren@superk.physics.sunysb.edu>
Date: Thu, 1 Jul 1999 18:13:06 +0900 (JST)
To: lwn@lwn.net
Subject: Difficulty of installation.

LWN mentions a Detroit Free Press article.  I requote your excerpt:

   "But I'm not going to be hard on myself, or Linux. No one buys a PC
   today without an operating system already installed. Putting
   Windows or the Mac OS onto a blank PC would probably be just about
   as difficult as installing Linux."

This got me thinking.  In my experience it is quite trivial to install
Linux on a blank PC (especially with our modern day distributions'
install methods).  What is the most challenging part is actually
installing Linux on what was previously a Windows only machine such
that Windows is kept around.

It is ironic that one of Linux's largest criticisms is due to trying
to play nice with one of Linux's largest competition (using that word
loosely).  

It would be nice if the folks in Microsoft would give as much
consideration to Linux and create an OS which doesn't, for example,
heavy handedly overwrite the MBR (LILO's usual resting spot) in order
to install.

-Brett.

From: nride@us.ibm.com
To: letters@lwn.net
Date: Thu, 1 Jul 1999 10:52:00 -0600
Subject: Assorted Benchmarks



In the midst of all the howling about the assorted benchmarks from Mindcraft and
PC Labs, I think people have missed something rather important. The benchmarks
are highlighting several areas where Linux could be better. We're getting a lot
of excellent information on how to make this OS faster and better. I propose
that we need more benchmarks, not less. We don't want benchmarks because we're
playing Microsoft's game. We want benchmarks to highlight potential kernel and
server issues. Issues that, once they come to light, will be fixed in a matter
of hours.

I propose also that we need to implement different types of benchmarks for
things that aren't normally measured today. Things like stability. We don't want
increases in speed if the trade off is a decrease in stability. So we need some
good stability benchmarks. Say, beat on the servers over the network and see if
they can survive. That sort of thing.

--
----------
Bruce Ide
nride@uswest.net

nride@us.ibm.com

Date: Fri, 02 Jul 1999 13:54:41 -0700
From: Tim Hanson <tjhanson@tscnet.com>
To: letters@lwn.net
Subject: Mindcraft

I take exception to your paragraph on the Microsoft / Mindcraft fiasco. 
Your introspection is admirable I suppose, but it brings to mind the
finding of purpose and meaning and important lessons to be learned after an
attack by a Great White.

Why do you suppose Microsoft advertizes for people who know Linux to lead
Linux strike teams, to make the best gosh darn products with which to
compete?  I don't think so.  It wouldn't surprise me that M$ knew all along
how Linux would do best case, with the guys from Red Hat doing the tuning,
and staged the whole thing, including the first tests to get us making
noise for them.  We just followed their script.

Maybe Linux advocates aren't up to dealing with one of this century's most
egregious commercial predators, the product of an individual agnostic to
any principle except maintaining his personal power over our society, Bill
Gates.  The Linux community, as it gets more under the skin of Gates and
those he promoted to positions of power, can expect to be baited like this
over and over again.  

Don't expect M$ to cooperate with any other benchmarks; Gates will never
use a deck he hasn't stacked.  All this discussion by Linux advocates about
the unfairness of the tests and the call for more balanced measures is
irrelevant.  Gates got to the audience he wanted; leaving us talking to
ourselves is no concern.  And that's where we're left.  Talking to
ourselves.

Date: Tue, 06 Jul 1999 14:03:59 -0600
From: Scott Marlowe <smarlowe@uswest.net>
To: lwn@lwn.net
Subject: lead page

I would have to take an exception to the statement made on your opening
page about the Mindcraft results.  You say:

BEGIN QUOTE

The Mindcraft rerun. The results are in; as expected, NT still beat
Linux strongly, though not so strongly as before. For this particular
set of tests, NT just performs better. For details, see the PC Week
article that first made the results available.

There are a few things to note about these results. First, perhaps, is
that much of the Linux community (including this publication) reacted a
little too strongly to the initial results. Certainly there were
numerous problems with how the first test was done, and it was right to
bring those to light. But, in the end, fixing the problems did not
change the ultimate results of the test.

END QUOTE

In fact, the results changed drastically in one regard: stability under
load.  While http accesses per second is a fairly worthless number to
grade a web server by, stability under load is a very important metric.
Apache/Linux as a web server was shown in the original test to be very
unstable under load, and worse yet, it did not recover after the load
was removed, it stayed in a crippled state and had the httpd server had
to be restarted to fix the problem.

The reason for this is simple, they compiled apache with the -O4 option,
when it is known to be unstable with any optimization over O2.  To this
day, Mindcraft has NOT retracted their statement in the initial survey
that showed Apache / Linux to be unstable.

I still hear from people who think Linux / Apache is unstable because of
this test, and your saying that the ultimate results did not change does
not help things.  We, the Linux / Apache community need to address the
issues raised by Mindcraft about performance, yes, but the issues of
stability have been almost totally ignored, when in fact, they are a
much more important point.  Linux / Apache is stable under heavy load,
even if a bit slow on a multi-NIC machine.

Date: Thu, 1 Jul 1999 15:32:49 -0400 (EDT)
From: Kristofer <kris@melon.org>
To: letters@lwn.net
Subject: GNU Cobol

> Converting Cobol to C would do nothing to improve the control flow of
> the programs, while making its data handling completely unreadable.
>
> Unless the only human resource one has available is C programmers, I
> would strongly discourage such a conversion.
>
> What the world needs is a free Cobol compiler - if this is the way to
> get one, even if it is not within the framework of the GNU Compiler
> Collection (which would make it retargetable to other architectures than
> the Intel ia32 model, among other benefits), then so be it.

Converting Cobol to C, while not improving the control flow, wouldn't hurt
it either, and the major advantage of C that justifies such a project is
that C already has a free optimising compiler for several different
architectures. If you don't want the C code, you can still do all your
work in Cobol, and forget about the time it spends as C code the same way
we all forget about how our C code spends time as assembly code when we
compile it. Instead of wasting our time reinventing the wheel, we should
be solving the real problem, which is giving the compiled software proper
debugging symbols that can be recognized by gdb as Cobol symbols, and
making sure the C layer is as transparent as the assembly layer.

Kris Coward,
Math Student/RA, SysAdmin
Univ. of Toronto

Date: Thu, 1 Jul 1999 15:41:23 -0400 (EDT)
From: Kristofer <kris@melon.org>
To: letters@lwn.net
Subject: MS Linux

> Amidst all the speculation and rumors about Microsoft coming out with
> their own proprietary Linux distribution, it seems that no one has
> considered a much more likely scenario. When the Red Hat IPO hits the
> market, MS could buy up virtually all the offered stock for $100,000,000
> or so (Bill could take it out of petty cash) and thus acquire a name
> brand Linux, not to mention the services of the Red Hat sales and
> service staff and their engineers and developers.

IIRC the shares of Red Hat being issued for sale will constitute only a
small ortion of the total wonershil of Red Hat.. certainly not enough for
MS to push any sort of agenda it may have.. of course, this doesn't rule
out their buying a whole lot of shares, they own a whole lot of Apple
after all.

Kris Coward
Math Student/RA, SysAdmin
Univ. of Toronto