Robert X. Cringely on the Apple / Intel deal

[Tony Hill]

Ever heard of Organizationally Unique Identifier?
http://en.wikipedia.org/wiki/OUI
http://standards.ieee.org/regauth/oui/oui.txt

<quote>
For Ethernet use, the OUI is combined with an internally-assigned
24-bit number to form a MAC address.
</quote>

Ethernet MAC addresses are, most definitely, unique.

No, they most definitely are not. There is absolutely no requirement
that the internal 24-bit numbers assigned by the companies are unique
and it is possible (and easy) to change a MAC address on many (most?)
ethernet devices.

 

[George Macdonald]

No, they most definitely are not. There is absolutely no requirement
that the internal 24-bit numbers assigned by the companies are unique
and it is possible (and easy) to change a MAC address on many (most?)
ethernet devices.

I'm not sure how the 48 bits are divided up but I think it's fair to say
that the original intent was that it be a universally unique number. It's
not that long ago that we had most ethernet cards with a MAC burned into a
ROM. Along the way some addresses were purloined by bottom feeders who
copied other mfrs' controllers & cards, including copying their burned in
addresses. As Keith has said, some mfrs are now recycling so the
modifiable MACs can be useful I suppose. OTOH I wonder how many orgs
bother to modify them and why?... what's the point? Are there now mfrs
selling cards where the burned in MAC address is repeated willy-nilly?

 

[keith]

I'm not sure how the 48 bits are divided up but I think it's fair to say
that the original intent was that it be a universally unique number.

That's true, but who would have thought that a 32bit number wouldn't be
enough for the Internet? The same thing applies. The fact is that
neither is a flat space and vast subsections were carved out for the
earlybirds. Even they have been recycling numbers.

It's
not that long ago that we had most ethernet cards with a MAC burned into
a ROM. Along the way some addresses were purloined by bottom feeders
who copied other mfrs' controllers & cards, including copying their
burned in addresses. As Keith has said, some mfrs are now recycling so
the modifiable MACs can be useful I suppose.

Not only purpose, but requirement. Ethernet is a tad old. I don't think
its designers knew where it was going.

OTOH I wonder how many
orgs bother to modify them and why?... what's the point? Are there now
mfrs selling cards where the burned in MAC address is repeated
willy-nilly?

There have been for at least ten years, maybe longer. The manufacturers
ran out of MACs long ago and have been forced to repeat them. As long as
two don't show up on the same subnet, who cares? If by some unlucky
coincidence they do, simply change one. AIUI, that's the reason they're
changable.

This shouldn't be a shock to anyone who's run the numbers. 2^24 is only
16M. Many manufacturers have made more than 16M Ethernet devices.

 

[Yousuf Khan]

I'd be very surprised if Intel didn't regard Microsoft's x86-64
decision as a declaration of war. That is to say, Microsoft already
_has_ optimized its code for AMD.

Why would it be offended? In the end, it saved Intel a lot of money
having to adopt AMD's instruction set, instead of trying to develop its
own from scratch. And Microsoft waited to release Windows until Intel
was ready to release its own chips.

And of course Microsoft has already optimized for AMD, who else were
they going to optimize for? It was there 1.5 years before Intel, and
much longer if you count pre-production chips which it would absolutely
have given to Microsoft first. Hell, even Microsoft's programming header
files call the platform "AMD64". Linux headers are actually more kind
towards Intel by calling the platform by the more generic "x86-64".

That would and should produce a shareholder revolt at Microsoft.

Sort of like the shareholder revolt that happened after Microsoft gave
cash infusions to Apple and Corel when they were looking shakey?

IBM's fab capacity is limited by its ability to make money making
chips, which it hasn't been doing in quite a while.

IBM doesn't seem to be able to produce anything with any reliability.
Nvidia was originally going to use IBM's Fishkill as its main source for
Geforce GPUs taking the job away from TSMC, but IBM had production
problems and Nvidia had to give it all back to TSMC.

Cray contracted IBM to produce the data router chips for its Red
Storm-line of supercomputers. Again, IBM had trouble producing enough of
them. To make matters worse, due to this delay, IBM itself was able to
get its own various supercomputers released earlier than Red Storm. Cray
has since gone on to give the job to TI.

IBM seems to not be very good at applying its own knowledge to its own
production. It's able to help people like AMD out with new production
processes, and AMD goes on to use it well. But IBM can't seem to improve
production in its own fabs.

All the feathers are already ruffled. There is simply no way that
Gates cannot be taking a dark view of the Intel-Apple deal, and Intel
is already furious over 64-bit decisions (x86/Itanium) by Microsoft.
The only question is where the bodies lie when the shooting stops.

Intel has nobody to blame but itself for its Itanium failures. As for
x64, it had nothing of its own, so it was likely going to use AMD's
stuff anyways.

Yousuf Khan

 

[George Macdonald]

There have been for at least ten years, maybe longer. The manufacturers
ran out of MACs long ago and have been forced to repeat them. As long as
two don't show up on the same subnet, who cares? If by some unlucky
coincidence they do, simply change one. AIUI, that's the reason they're
changable.

This shouldn't be a shock to anyone who's run the numbers. 2^24 is only
16M. Many manufacturers have made more than 16M Ethernet devices.

But a full MAC address is 48 bits. Is half of it reserved for some
Ethernet thingy?

 

[Robert Myers]

Why would it be offended? In the end, it saved Intel a lot of money
having to adopt AMD's instruction set, instead of trying to develop its
own from scratch. And Microsoft waited to release Windows until Intel
was ready to release its own chips.

And of course Microsoft has already optimized for AMD, who else were
they going to optimize for? It was there 1.5 years before Intel, and
much longer if you count pre-production chips which it would absolutely
have given to Microsoft first. Hell, even Microsoft's programming header
files call the platform "AMD64". Linux headers are actually more kind
towards Intel by calling the platform by the more generic "x86-64".

Did you miss the episode of Intel trying to get Microsoft to implement
Windows for an incompatible Intel 64-bit x86 extension?

We seem to be off into speculation about what nice people do in a nice
world. Intel is unhappy with Microsoft over Microsoft's 64-bit
decisions, and it doesn't make the slightest difference to Intel's
attitude toward Microsoft what Linux calls x86-64.

Sort of like the shareholder revolt that happened after Microsoft gave
cash infusions to Apple and Corel when they were looking shakey?

Intel spent that money in an attempt to obfuscate the fact that it
*is* a predatory monopoly.

Intel has nobody to blame but itself for its Itanium failures. As for
x64, it had nothing of its own, so it was likely going to use AMD's
stuff anyways.

The only reason Intel implemented x86-64 and not some proprietary
variant is because Gates left them no choice. Whether Intel has only
itself to blame or not for itanium failures doesn't matter. Intel
can't be happy at Microsoft decisions to support x86-64 and not
Itanium.

RM

 

[Robert Myers]

Intel spent that money in an attempt to obfuscate the fact that it
*is* a predatory monopoly.

You'll never let that one go. It should read that Microsoft spent the
money to obfuscate the fact that *it* is a predatory monopoly. As to
Intel being a predatory monopoly, it would like to be, I'm sure.

RM

 

[Kai Harrekilde-Petersen]

I'm not sure how the 48 bits are divided up but I think it's fair to say
that the original intent was that it be a universally unique number.

It's a 24-bit OUI and a 24-bit serial number. At least on Ethernet,
bit 40 og the OUI is the multicast/unicast bit, and bit 41 is the
Local/Global bit.

Regards,


Kai

 

[AD.]

The fact that they can be changed (easily) and are not FORCED to be unique
means that you can most definitely not *depend* on them being unique, ie
trying to design some sort of DRM scheme around MAC addresses is an
exercise in futility.

Agreed. I just misread your statement to mean something like "they are by
design unique on a subnet but not unique globally".

Not having read the article (whadya you mean this isn't Slashdot?), I
didn't realise the context was around their usefulness in DRM schemes.
That little extra bit of context makes your meaning much clearer now

 

[AD.]

And will Apple PCs using Intel chips also run Windows?

Probably IMO.

Will Apple SELL PCs with Windows?

No chance. Then they'd have to support it like all the other OEMs. I can't
imagine Apple being willing to do that.

Then again when 'support' these days consists of "pop in the recovery CD
and reboot", maybe they would?

 

[chrisv]

There have been for at least ten years, maybe longer. The manufacturers
ran out of MACs long ago and have been forced to repeat them. As long as
two don't show up on the same subnet, who cares? If by some unlucky
coincidence they do, simply change one. AIUI, that's the reason they're
changable.

Hmm... They're not customer-changable on the products my company
makes...

 

[George Macdonald]

It's a 24-bit OUI and a 24-bit serial number. At least on Ethernet,
bit 40 og the OUI is the multicast/unicast bit, and bit 41 is the
Local/Global bit.

Ah, OK - now I see... there's even a list of OUI "known" mfrs:
http://standards.ieee.org/regauth/oui/oui.txt. Certainly I can see that,
now that mbrds can have on-board NICs -- and some more than one -- where
they are going to run out of serial numbers... and maybe quite quickly. I
also wonder about all the OUIs which are locked up through mergers etc.
over the years being badly distributed.

 

[keith]

Hmm... They're not customer-changable on the products my company
makes...

Really? They've been changable on everything I've worked on for years
(and no, I've not had any reason to look recently). The utility to change
them (it's in the serial EEPROM) may not be obvious, but it's there.

 

[keith]

Ah, OK - now I see... there's even a list of OUI "known" mfrs:
http://standards.ieee.org/regauth/oui/oui.txt. Certainly I can see that,
now that mbrds can have on-board NICs -- and some more than one -- where
they are going to run out of serial numbers... and maybe quite quickly. I
also wonder about all the OUIs which are locked up through mergers etc.
over the years being badly distributed.

I believe we've had the same argument here over IP addresses. 64bits
should be enough for *everyone*! ...except that reality steps in.

I still have a hard-wired IP address (perhaps two) out of the 32bit space.