linuxmafia.com 2016/linuxmafia.com.tar/linuxmafia.com/pub/hardware/bluenotes-editorial-1996-08

Text File | 1997-01-01 | 13KB | 229 lines

Some months back, in my compendium of on-line Microsoft humour (the one
kicked off by the now-classic gag "press release" about Gates buying the
Catholic Church), I mentioned the view, widely held on the Net, that the
computer press is corrupt and in Microsoft's pocket -- and said that I
disagree. I didn't elaborate at that time, but I will now.

It's easy to see why so many feel that way: To users with broad
computer experience, the press's obsession with even the most vapourous
Microsoft products and initiatives has seemed ludicrous, for quite a
few years. One leading example is the ongoing push for Windows
NT, a slow, bloated, none-too-compatible product whose market share
still remains minuscule, even after its much-ballyhooed, three-year-late
introduction, several years ago. (When the initial versions rather
stunningly flopped, magazines such as <I>PC<D>, <I>PC World<D>,
<I>InfoWorld<D>, <D>ComputerWorld<D>, <I>Byte<D>, and so on showed only
minor embarrassment over their protracted gushing, but
then promptly forgot the experience, and resumed their drumbeat a year
later.)

Yet, when accused of favouritism, those magazines' writers and editors
very heatedly and consistently deny the charge -- and cite
copious evidence of their independence from advertisers. I believe them:
As I see it, the truth is much less dramatic.

Last year, there was an intriguing panel discussion on this very
topic, at a meeting of the OS/2 Bay Area User Group. Quite a few
of the most famous industry editors and writers were kind enough to
participate. The main question was why OS/2 (like other non-MS
operating systems) gets such disproportionately tiny, ill-informed
coverage, and how this might be fixed. Several panelists gave the same
answer: IBM simply needs to promote its offerings a great deal better.

Now, sometimes I'm a little slow: At first, this struck me as a
<I>most<D> peculiar answer. Why should the accuracy and fairness of
<I>InfoWorld's<D> coverage depend on IBM's marketing? Isn't that a non
sequitur? (Don't these magazines purport to research the facts, and then
report them?)

Well, so it seems, unless you consider how the computer press in
fact works. Most of the magazines are run by overworked staffers on tight
schedules. The central offices' computers are rarely sophisticated as to
either hardware or software, and many of the feature pieces are by outside
contractors of uneven quality. Given the constant demand for quick
production, any text or other information from vendors (such as
Microsoft) is a godsend -- particularly in a quickly-moving
technical field where most reporters are writers first, and
technologists second.

Writers and editors are constantly plied with press releases,
backgrounders, white papers, and pre-release samples. They're also
whisked off to weekend-long "technology conferences", where vendors try
to prime the PR pump by supplying them with prefabricated coverage
in the form of press kits. (You can predict the press's party line
about a month in advance, by following such events.)

Microsoft is, of course, the master of press relations, making
concentration on what it promotes the path of least resistance.
That's a formidable advantage.

In short, that wasn't a non-sequitur answer, because of the fact -- seldom
discussed because it's an awkward truth -- that these reporters, even
as capable as many of them are, really do <I>very little reporting<D>.
Rather, they (mostly) reflect what they are fed. Nods are occasionally made
towards "balance" -- usually in the form of printing two readers'
letters with opposing views. The subtext is clear: If you want them
to print fairer coverage, then make sure they're <I>sent<D> some.

An uncharitable interpretation would be that the industry prints mostly
regurgitated press releases -- biased in favour of whoever has the best
PR machine -- because it's lazy, and doesn't care about the readers'
interests. My view, though, is that this is all that the readers demand
and care to pay for, so it's all they get. Most readers are so
uncritical, these days, that they might not know the difference.

[********SUBHEAD FOLLOWS********]

Still Chewing the FAT

Speaking of not knowing the difference, another piece of half-baked
technology from Microsoft has just emerged: Windows 95 OEM Service
Release 2. Contrary to misinformation conveyed at our July general
meeting, this revised version is <I>already<D> shipping to OEM vendors
(such as, say, Gateway 2000), and appearing on new computers as
pre-loads. This is an OEM (Original Equipment Manufacturer) upgrade
<I>only<D>. Again, contrary to what we heard at the general meeting,
there are no plans for its release to the general public.<^>1<D> It is
available as an OEM preload, only (and should not be confused with
the earlier small package of bug fixes, Service Pack 1).

How do you know you have it? Open a DOS session (eek!), and type "ver
/r". If MS-DOS 7 reports back the version as "4.00.950", it's regular
Win95. If it says "4.00.1034", though, you have OEM SR2.

So, what's the deal? It adds native support for removable IDE drives,
high-capacity "flopical" disks, IDE busmastering, and a few multimedia
extensions. There's a new (and less gargantuan) e-mail client, replacing
Exchange, and some more-stable support for IRQ-sharing (gaackk!).
Mostly, though, people will notice a new, incompatible disk format,
"FAT32".

FAT, as readers of this column may recall, is DOS's one and only disk
format, a creaky invention copied bodily, 20 years ago, from the 160 KB
(no-subdirectory<^>2<D>) <I>diskette<D> format used by Digital Research's
CP/M operating system. (Contrary to the claims of Microsoft's copious
press releases, FAT was <I>not<D> invented by Bill Gates, nor by anyone
else at Microsoft. This CP/M-clone format was acquired when Microsoft
bought -- not invented -- DOS from Seattle Computer Company.)

FAT is named for its primary feature, a File Allocation Table of 65,536
(2 to the 16th power) housekeeping entries that each keep track of a
"cluster" (one allocation unit) of storage. At the time one formats a
hard drive volume, DOS decides how big a "cluster" must be -- how many
512-byte "sectors" -- in order that the 65,536 available entries might
span the entire volume.

Now, this was a perfectly fine scheme back in the days of <I>160 KB
diskettes with no subdirectories,<D> but a horrid way to run, say, 2 GB
hard drives. Each FAT entry has to figuratively wear seven-league boots,
making DOS unable to allocate storage in dollups smaller than 32 KB on
larger volume sizes, even if storing only a 1 KB file. The waste of
storage can be incredible: up to 30% or more of total drive space.

This was a glaring problem, but <I>not<D> the only one. FAT volumes
rapidly fragment both files and directories, requiring use of utilities
to disentangle them (a DOS-world fetish unknown on OSes with
better-designed file systems, including Macs). The more fragmented they
become, the more failure-prone data integrity measures such as
un-erasure of files also become. Worse, performance is so bad that the
OS has increasingly compensated for it using delayed-write caching,
increasing the FAT's vulnerability to damage in case of power
interruptions or crashes.

With Windows 95, Microsoft said it had a "new format", called VFAT,
promising all sorts of wonderful advances. On examination, though, it
turned out to be just plain old FAT, with a kludged set of hidden
regular directory entries stitched into it, to store long filename
information -- which silently has the effect of worsening FAT's
fragility and fragmentation problems.

That brings us up to the present, with Win95 OEM SR2 and FAT32. What
about it? First of all, the name is misleading. Contrary to Microsoft
literature (and its reflections in <I>PC<D>, <I>InfoWorld<D>, etc.),
it's not "32-bit" anything (a much-abused term that should be given a
vacation): FAT32 uses 28-bit entries in its table, with a bigger boot
record per volume, and new partition types in the partition table
(thereby, by a curious coincidence, breaking OS/2's Boot
Manager).<^>3<D> Also, the root directory becomes no longer fixed in
size or location, there's space for sundry new data fields (free space
per volume, etc.).

The 28-bit FAT means there can be 2 to the 28th power or 268,435,456
entries in the FAT table (up from 65,536), which in turn means those
entries can manage very small (even 512-byte) clusters, when you format
even very large volumes. Smaller clusters make possible huge FAT volumes
without cluster-related wastage, which is good, right? Well, yes and no.

A tremendous number of very small clusters also means more overhead
managing them all (including more RAM for in-memory copies of FATs,
<I>and<D> it means greatly accelerated fragmentation. Further, Win95's
somewhat hyperactive virtual-memory driver does <I>cluster-based<D>
read-aheads on your hard drive (it reads a certain number of clusters
past your present needs, in case adjoining data will be needed), so the
swapper may become less efficient. (I haven't had time to test this.)

However, as the old line goes, "But wait, there's more!" A fair
cross-section of the following types of utilities are going to break
completely, leaving you with the fun of re-buying them (if new versions
become available): hard disk repair utilities, unerasers, anti-virus
packages, disk compression (e.g., Stacker), tape backup, security
programs (e.g., Norton DiskLock), uninstallers, games and other
applications that use copy protection, defragmenters (!), and disk
drivers (e.g., ASPIDISK.SYS). If you're lucky, the broken ones will
merely crash. If not, they may start writing abstract designs atop your
data. . . .

. . . and all this for <I>what?<D> Even with the antique FAT/VFAT format
that DOS/Windows users have endured until now, tolerable results have been
possible with intelligent partitioning and occasional disk maintenance.
This new thing brings only a small space savings over careful "FAT16"
setups, at the cost of even worse performance, worse fragility, and a
veritable junkyard of broken utility packages.

Back when Win95 and its <I>billion<D>-dollar promotional campaign hit
the streets, I remarked that it's a crying shame they didn't seize this
golden opportunity to offer a <I>real<D> file system instead of FAT.
The same goes double with OEM SR2.

If you check the Microsoft PR Department's explanation
(http://www.microsoft.com/windows/pr/fat32.htm, with subsequent articles
from <I>InfoWorld<D> et al. bobbing in Microsoft's wake, as usual),
you'll see that they dismiss what is proclaimed to be the <I>only<D>
alternative, WinNT's NTFS disk format, as unwieldy. However, as usual,
they omit a more obvious option: the much better-performing HPFS format,
to which Microsoft still has rights from the days of its partnership
with IBM, and which works at blazing speeds in OS/2 2.0 and above. (It's
also used on Novell NetWare servers to support -- ironically -- Windows
95 long filenames.) HPFS is fast, is damage-resistant, holds long
filename information without ugly VFAT-type kludges, and
self-defragments.

Why didn't they support it? The reason may be that it'd be embarrassing
to so openly adopt IBM's technology, instead of their own. Also,
Microsoft seems to still be "accidentally" impairing interoperability
with OS/2 at every turn: HPFS support will be dropped in the next
version of NT, where it has until now served as a higher-performance
alternative to FAT and NTFS on workstations<^>4<D> -- and there's also the
aforementioned blow to OS/2 Boot Manager.<^>5<D>


<^>1<D>

<^>3<D>FAT32 will not work with Boot Manager at all. This is a step
beyond Microsoft's mere gratuitous deactivation of Boot Manager every
time one runs the DOS 6.x, Win95, or NT installer. In the latter case,
users quickly learned to simply re-install Boot Manager thereafter.
This new step is more of a challenge.

<^>4<D>

<^>5<D>I'm not sure why this is, but it may have something to do with
two new type numbers Microsoft puts in the partition table (0B hex for
regular FAT32, or 0C for FAT32 with LBA translation). Anyhow, for those
wanting multiple boot partitions while waiting for IBM to compensate,
there are alternatives: (1) FreeBSD's BootEasy will boot anything, and
puts up a text screen with your choices selectable using function keys.