response 28 of 136:

Mar 31 03:18 UTC 2002

The PPC market is split - there's IBM, and then there's apple.  There's
also Motorola and a few speciality folks.  IBM makes some very nice
higher end stuff, which is only mildly overpriced.  Apple has
traditionally *never* included ECC or even parity.  Every organization
has its corporate blind spots.  Apple apparently feels it's more
important for something to keep running than to detect and report
errors.  (Contrast this with the original IBM PC which included parity
in an era when personal computers *never* had parity, because they came
from a background (banking/business) where errors are simply not
acceptable.)  Motorola and the speciality folks make machines for
various market speciality market niches, such as engineering, embedded
controllers of various sorts, & so forth.  For a bit, it looked like
IBM, Apple, & Motorola were coming together to make up a common
motherboard standard.  If this had come off, there might have been
enough mixing of these various markets to make it possible for grex to
purchase a low-cost motherboard with the appropriate features.
Unfortunately, this never happened, which is regretable on many levels.

Technical arguments, with their basis in engineering and mathematics,
are even more prone to schisms based on original premises.  Any engineer
who can't admit this is so hopelessly tied up in his original premises
that his judgement can no longer be relied upon, in any field.  And, in
fact, in the previous paragraph, it's easy to recognize different
rational engineering decisions being made based upon fundamentally
different initial premises.  Pure science is no less immune to this
disease of the initial premise.  Most of the more important scientific
discoveries are made by relatively young people, and often by people who
aren't in the mainstream.  It then generally takes about a generation
for the discovery to become generally accepted, which is oddly enough
the time it takes for the old establishment to die out.

Sure, we want a fast machine, but we also want a *Consistent* machine.
I've heard from various people, at various times, that PC hardware
generally tends to behave more inconsistently and less pleasantly, under
great load.  What that would translate to in human terms, is a machine
that is faster by any objective measure, but feels subjectively slower
and is more annoying to use.  Measuring pure speed is easy to do.
Measuring *consistency*, load, & human factors is not nearly as easy.

Also, lest people claim I'm immune to the intel argument, allow me to
point out: I have intel hardware at home (well, actually that and AMD
K6-2).  STeve is still very proud of his pentium 4 based IBM laptop,
with its 1600x1200 display and running OpenBSD.  At work, I may be
switching our mail server over from running on an IBM RS/6000 530H, to
some newer intel based thing.  Different hardware and software are best
at different things; there is no one best answer for all things.

response 30 of 136:

Apr 1 01:38 UTC 2002

Regarding the initial premise thing, it strikes me that if I have a
microcontroller running a safety critical pressure release valve, it
damn well better detect and report errors.

I've never heard this inconsistency argument before, and I've been
running Unix on x86 hardware for many, many years.  Perhaps you could
be a little more specific?

response 32 of 136:

Apr 1 20:45 UTC 2002

Really?  I don't know; I find Sun's feel uniformally slower than x86
machines, even under really heavy load.  The Sun I'm sitting in front of
right now feels extremely slow, and the number of processes running on it
isn't that great.  Hardware multitasking is nice, but really no substitute
for good context switching code.  An empirical piece of evidence is
Plan 9 running on x86 hardware compared to running on, say, SPARC.
It's just not faster, no matter what kind of load is running on it.

What we're really talking about is scalability.  x86 hardware really does
scale as well as other stuff these days, particularly at the level we're
talking about here.  Witness that the main Plan 9 CPU server at Bell
Labs is a quad-processor Xeon machine.  I can personally tell you that
that machine (which replaced an 8-processor SGI Challenge XL machine)
is hammered quite frequently, with both big compute jobs and lots of
processes.  The hardware context stuff just isn't needed, but Plan 9 is
probably a lot more efficient than Unix at doing context switches.

response 34 of 136:

Apr 10 13:39 UTC 2002

Right.  The intent isn't to use it as a production machine, but
rather as a testbed for OpenBSD on x86 hardware.

response 36 of 136:

Apr 10 19:43 UTC 2002

Wow, that's cheap.  Grex will likely want SCSI, which adds a bit of cost,
but only a few hundred dollars, and there's money in the bank specifically
earmarked for hardware.

response 38 of 136:

Apr 11 16:28 UTC 2002

ECC memory is cheap.  I've seen gigabytes of it for sale on the streets
of New York for under $100.  It's really not that hard to put together a
PC that's a steal that still has the features you want.

response 40 of 136:

Apr 12 06:03 UTC 2002

No wonder ECC memory is hard to get; it's only available from NYC street
vendors.

response 42 of 136:

Apr 12 16:58 UTC 2002

ECC would have been $100 more on my new system.

response 44 of 136:

Apr 14 14:20 UTC 2002

But MDW is working on something to do with Dan Cross's donated computers?
It sounds like whoever is willing to do the work gets to choose the system.

response 46 of 136:

Apr 14 15:41 UTC 2002

It's not?  It sounds okay to me.

response 50 of 136:

Apr 15 03:00 UTC 2002

So who is volunteering to do what Jan is now too busy to do?  Shall we wait
until Arlo is old enough to set up a new system by himself?
The most appropriate tool may be the one that there is someone willing to set
up and maintain.  Has anyone besides Marcus volunteered recently?