[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Ultrasparc 3



From: Chuck Yerkes <chuck+obsd_(_at_)_2002_(_dot_)_snew_(_dot_)_com>
> Date: Mon, 25 Nov 2002 22:21:53 -0800
> From: Chuck Yerkes <chuck+obsd_(_at_)_2002_(_dot_)_snew_(_dot_)_com>
> To: misc_(_at_)_openbsd_(_dot_)_org
> Subject: Re: Ultrasparc 3
> Message-ID: <20021125222153_(_dot_)_B29037_(_at_)_snew_(_dot_)_com>
> Reply-To: misc_(_at_)_openbsd_(_dot_)_org
> 
> > Sun apparently lives, eats, and breathes NDAs.  That's how they do
> > business, and I think they fail to see at a corporate level how that's
> > just so fundementally incompatible with opensource.
> 
> Sun has always been a hardware company that has strong ties
> to their software side.  They often hold their information
> close to the chest.  It's how they run.

True, and true.  But that's not the complete picture.  Actually, with
the SUN-1, you *could* get complete information.  That's because Sun
Microsystems did not exist.  I have at home a very nice schematic for
the SUN-1 cpu board, which came with a Codata CTW-300.  It's true, SUN
went into paranoid mode with the SUN-2.  But, it still used multibus.
It used the 68010, on which information was readily available.  The
stuff inbetween was still close evolutionarily speaking to the SUN-1.
It was easy enough to recreate the missing information so that didn't
really hurt sun.  Even with later hardware, it was often still possible
to buy a license to sunos source, and see how the software driver
worked.

> 
> Digital had different divisions that they let run (and nobody
> would market).  The Alpha division was expected to try to be
> profitable.  If that meant selling Alpha's with OSF/1 or VMS,
> fine.  When OS people (NetBSD, Linux folks) came to them for
> information, well, they want to sell Alpha chips and boxes so
> fine.  And the community got pretty good disclosure and detailed
> information.  I recall NetBSD having some really good docs on
> a (mips) DECStation that they, er, "backed up in case DEC
> forgot that it was okay to publish."

Alpha opened up later.  I think that was after they discovered
fierce customer resentment.  Unfortunately, I think it was
also after the damage had been done.

> 
> Sun gains pretty much nothing by helping OpenBSD out.  OpenBSD
> doesn't have SMP; Sun hardware is high priced and pretty useless
> for single CPU machines.  The best use is to have something
> to build software on for your big machines.  The Netra's are
> about the only compelling box out there, and that's pretty
> much just due to having a nice and smart serial console.
> (oh, for a real PC BIOS that spoke serial if no keyboard)

I disagree.  Well not about the PC BIOS thing - that's a real
mess which I expect to learn with some pain.

On my wishlist for openbsd is SMP and "kernel" threads.  Whether I'll
get to it is another question, but at least I have the hardware (sort
of): I now have an ultra-2 sitting at home with 2 ultra1 cpus.
At work, we run AFS file servers - being able to have one process
do multiple disk reads *and* service incoming packets is a real
win.

New machines are expensive, but used machines are much more
reasonable.  I also do volunteer work for grex
(http://www.cyberspace.org/).  There sun hardware is attractie: we
provide free unvalidated shell access to unix.  Not running the intel
instruction set isn't really any more secure, but it does stop 98% of
script kiddies (we call 'em "cookbook vandals") just fine.

I also disagree that this doesn't help sun.  Sure, it doesn't sell any
machines *directly* today.  But: sun's market is research/education.
Somewhere out there somebody today is writing the answer to Mach,
Windows, and all else -- the OS we'll all wish we were running 10 years
from now, while kicking ourselves that we didn't think of the idea
first.  In the 1970's, such things got written for the pdp-11.  In the
1980's, they were vaxen, then 68k.  Nowadays, it's all i386.  Sun
needs to give them a reason to think about something besides
i386 or ppc.

Another thing: sun is indeed hurting for money.  They have inhouse a
huge OS development effort, and a smaller compiler team.  In both
cases, these people are doing what Sun *might* do less expensively by
going opensource.  Already, there are components (apache, perl,
sendmail, maybe X, and as you note, gnome) for which Sun is in part
already dependent on opensource.  I'm not sure if they yet have openssh
- but if they did that makes them directly dependent on openbsd.  I
don't expect to see sun lay off all their kernel hackers & compiler
writers - but I would expect them to find ways to free people up for
other projects and reduce their internal support costs by relying more
& more on externally produced opensource software.  Ie, get more bang
for the buck.  As a hardware company, this ought to appeal to them.

...

					-Marcus Watts