See also: last week's Back page page.

Linux Links of the Week

In case you've not looked at enough patent information...what was once the IBM patent database is now Delphion. Should you ever want to delve into the details of a patent, this is the place to look.

Section Editor: Jon Corbet

This week in history

But what's really held IBM back from an official support alliance with Red Hat, say sources close to the company, are legal issues. If IBM supports Linux the way it supports other operating systems, it will need to tweak the operating system itself, and that could raise liability questions neither IBM nor its many partners want to deal with.

People seemed to have reached a point where they worry a lot less about the liability issues.

ZDNet ran a Top Tech Newsmaker poll. Linus Torvalds came in second, having been beaten, 2-to-1, by Jenni of the JenniCam.

After resisting for some time, Red Hat quietly dropped a set of KDE RPMs into its "Rawhide" distribution. Red Hat also put an end to its practice of dropping updates into second and subsequent pressings of its CDs. Until then, one Red Hat 5.0 CD could be visibly identical to another, but have a different set of packages.

GNOME 0.99.0 was released.

One year ago (December 23, 1999 LWN): Eric Raymond announced his forthcoming book, The Art of Unix Programming. The book was to document what makes the Unix tradition special, and was to be written with a great deal of help from the community. Eric did not, however, set a deadline for this book; currently it is available through Chapter 2.

People wondered about the 2.4 kernel...

Colin Tenwick, vice president and general manager European operations for Red Hat, confirmed that the kernel would be released formally to the Linux community the same time as Windows 2000.
-- VNUnet, December 20, 1999.

Needless to say, things didn't happen that way. In an attempt to get a guess at when the release would happen, Tummy.com announced its When's 2.4 poll. From the results page we see that the most optimistic entrant picked January 27, 1999. A full 87.7% of the entries taken so far (it is still open, obviously) gave dates prior to the present.

Richard Stallman called for a boycott of Amazon.com as a result of Amazon's use of software patents.

Linux-Mandrake 7.0 beta was released, as was Mozilla M12.

Corel's Linux distribution was due to hit the shelves any day. Meanwhile, the company foreshadowed the general decline in Linux stocks by dropping down into the low teens from its high of $43. Of course, the low teens would look pretty good to Corel investors these days...

Red Hat, instead, announced a two-for-one stock split.

Even if Linux does turn out to be the greatest thing since the graphical user interface, I sincerely doubt that people buying shares of VA Linux (or any of the Linux companies) at their current valuations will do anything but lose sleep and/or money.
-- Pat Dorsey, Morningstar.

Letters to the editor

Date: Fri, 15 Dec 2000 00:51:21 +0100 (MET)
From: David Kastrup <David.Kastrup@neuroinformatik.ruhr-uni-bochum.de>
To: letters@lwn.net
Subject: Defamation of Pascal


I take exception to the uncommented and uninformed bigotry of Linus
Torvalds against Pascal in his one-year old quote given in last Linux
Weekly, especially as this nonsense is picked up by the editor:

          The fact that our forefathers were Pascal-programmers, and
          started counting from one does not mean that we have to
          continue that mistake forever. We've since moved on to C,
          and the change from 1999->2000 is a lot more interesting in
          a base-10 system than the change from 2000->2001.

    Of course, it looks like no 2.4.0 by the end of the millennium
    even by the reckoning of Pascal programmers...

The pure unadulterated fact is that Pascal is agnostic.  Its arrays
start wherever the programmers want them to start.

You can declare:

var at1: array [1..30] of integer,
    at0: array [0..29] of char,
    at5: array [5..10] of -3..7;

Pascal has enough faults of its own.  We don't need to start
phantasizing about imaginary ones.

Most probably Mr. Torvalds is confusing Pascal with Fortran which
indeed has its arrays starting at 1.

David Kastrup                                     Phone: +49-234-32-25570
Email: dak@neuroinformatik.ruhr-uni-bochum.de       Fax: +49-234-32-14209
Institut für Neuroinformatik, Universitätsstr. 150, 44780 Bochum, Germany

Date: Wed, 20 Dec 2000 16:21:53 -0500
From: "Jay R. Ashworth" <jra@baylink.com>
To: letters@lwn.net
Subject: "Network Appliances" -- what a difference a year (and Linux) makes

Well, tis almost Christmas.

One of the presents a manufacturer hopes you'll put under someone's
tree this year is the New Internet Computer; the Internet- and Linux-
based network computer from a subsidiary of Oracle -- the company
that's been trying to promote the network computer idea for years, in
hopes of breaking the Microsoft monopoly; Larry Ellison reputedly hates
Bill Gates' guts.

The rapid advances in Linux development in the last several years,
combined with the near-ubiquitous presence of the Internet in the
American home, and the appearance of motherboards that actually
have integrated peripherals powerful enough to *use*, have brought
things to a point where the New Internet Computer Company can sell
their "Internet Appliance" for $199, without a monitor but including
everything else, *without* an Internet bundling/kickback deal.

This last point is very important: has anyone heard anything of the...
damn; I can't even remember the name :-) -- that integrated LCD screen
IA that was released around last Christmas, made tandem waves by being
hacked, and then by annoying the manufacturer because it was being
hacked.

Why was the manufacturer annoyed?  Well, because they decided on a
"give away razors; sell blades" business modem that had them making
most of their money off the bundled Internet access service... which
the hackers weren't using.

The NIC people didn't make this mistake.  You can use their little
black box with any internet service you like; commercial dialup
services, Netzero, even BAMnet, a Pennsylvania based combination ISP/
CLEC who bill the long distance and access at a combined 6.5c/minute,
passing it through to your local phone company; no advance setup,
apparently, required.  (Presumably, you have to dial that carrier's
1010 code, but that's it.)

Even more importantly, though (to me, and I suspect to many others)
will be the fact they've *also* included 10/100 Ethernet (and USB) on
the box.  The theoretical advantages of the X-terminal model in
business have long been understood -- minimization of desktop
maintenance and the like being the biggest -- but the things have just
been too damned expensive.

Not anymore.

$200 plus a monitor is sufficiently cheaper than a "real PC" to make
the *inital* cost differential something you can sell to management,
let alone the reduction in recurring costs.  The software is Linux
based; the changes for the custom hardware are GPL'd; the application
field for Linux is growing... how could it get much better?

I've talked to NIC's sales manager, concerning the one point in their
sales agreement that gave me pause: they effectively limit sales to
retail end-customers.  He tells me that this is primarily because their
margin on the boxes is razor thin -- they won't be offering discounts
at *all*.

Hey, that's ok for me.  I can add enough value to sell them for $300 a
desktop, and still save people a shitload and a half of money in the
long run.

And since they got all the glue right, they'll be useful to many
people, in many environments, that the original manufacturer (in the
immortal words of Arlo Guthrie) hadn't even counted upon.

In the long run, since they make, not lose money on every unit, that
can only benefit the manufacturer.  Case in point example to everyone
else.  In the Internet Age, you can't *pull* a fast one; it will break.

[ checks the LWN archives ]

Let this be a lesson to the manufacturers of the i-opener; perhaps
their choice of product name will prove more prophetic than they had
expected.

A quick glance at their website ("you've tripled the price, why?" and
"Why is Netpliance announcing a 'membership program'?") suggests not.

I especially like the *answer* to that last question:

> Netpliance is introducing the Membership Program because the components
> of the program will provide our customers with an even more fulfilling
> online experience that, once again, is centered around simplicity. This
> is the principle upon which we founded the company. We choose to
> reinforce that commitment to our customers whenever we can.

Does anyone here need any more on that?

Cheers,
-- jra
-- 
Jay R. Ashworth                                                jra@baylink.com
Member of the Technical Staff     Baylink
The Suncoast Freenet         The Things I Think
Tampa Bay, Florida     http://baylink.pitas.com                +1 727 804 5015

From: Anton Ertl <anton@a0.complang.tuwien.ac.at>
Subject: Elevator algorithms
To: letters@lwn.net
Date: Thu, 14 Dec 2000 14:55:59 +0100 (MET)

There are a few aspects that earlier messages missed:

All file systems require that the disk writes the blocks in the given
order, otherwise the file system may be corrupt upon a power outage
(and yes, we have had power outages on system with an UPS, due to the
UPS breaking down).  That's certainly true for journaling and
log-structured file systems, and file systems that use ordering
constraints to maintain a certain amount of consistency (e.g., BSD FFS
without and with soft updates).  Even file systems like ext2 that rely
on fsck instead of write ordering constraints, need synchronous writes
upon fsync.  Data bases also have write ordering constraints (they
usually work similar to journaled file systems).

AFAIK the only way to ensure these write ordering constraints are
heeded on current disks is by turning off write caching (see
http://www.complang.tuwien.ac.at/anton/hdtest/).  So anyone with
valuable data will turn off write caching; on a SCSI disk I measured
there was no way to enable write caching, presumably because SCSI
disks are usually used for valuable data.

Unfortunately no-write-caching implies a total order on the requests,
whereas the actual write order requirements are usually a partial order.
An elevator algorithm can exploit the freedom available in the partial
order to produce a more efficient sequence.

- anton