[Ti] Apple's True Market Share!

[Mark C. Langston]

On Thu, Dec 12, 2002 at 12:52:44PM -0600, Chris Olson wrote:
> Mark C. Langston wrote:
> 
> > That's not entirely accurate -- there is a private beta period.  Early
> > Access is closer to a gamma test than beta test.  Still, this is a
> > normal part of their development cycle.  There will be a final version
> > of OS 9;  this isn't it.  It does, however, contradict the previous
> > poster's claim that Sun has given up on x86 development.
> 
> We are a Sun support center, and I beta test everything that comes down 
> the pipe.  Solaris on x86 is not a fully funtional, production 
> environment operating system, nor will it ever be.  It is a 
> learning-point port only, for those who can't afford, or won't buy the 
> native hardware.  It has the problems that Apple would have if they 
> tried to port to x86 - a specific HCL (look at Darwin's HCL), and very 
> poor performance due to the inferior x86 processor design and it's 
> associated instruction sets.  If your hardware is not in the HLC, forget 
> it - it won't work.  As I also already posted, supporting all the 
> hardware on the x86 architecture is a nightmare for a truly 
> high-performance operating system.  There are major, major differences 
> in RISC vs. CISC architectures.

I've also been in various positions in a company providing Sun support
(heck, we even designed portions of their certifications), from techie
to CTO.  I've also personally been a beta-tester for various versions
of Solaris (2.6 and 7, respectively.  I've been working professionally
with Sun operating systems since SunOS 4.1.1, in large-scale production
enviornments.)

What, exactly, do you mean by "not fully-functioning"?  All the daemons
work just fine; the scheduler schedules.  The slicer slices.  The
memory manager manages.  Now, if you want to complain that the graphics
are slow, so be it -- When I hear someone complain that a Unix doesn't
work, I start looking for basic operating system problems, like a
poorly-written kernel, broken /dev trees, and the like.  I don't stop to
think, "wow. This person means their pretty pictures aren't."  As far as
I'm concerned, previous versions of Solaris x86 work just fine.  That is
to say, their functionality is indistinguighable from the outside, and
only marginally so when working on one as a sysadmin.  If you're using
both as desktops and complaining, then sure, x86 sucks.  But then, so
does the SPARC version...Sun's marketing department to the contrary,
Solaris is not a good desktop OS, and was never designed to be.  The few
attempts they've made towards it are either promptly disabled by
competent admins, or replaced with better-beahved and better-written
equivalents. 


I fully expect SunOS 2.9 (nee Solaris 9) on x86 to be similar.

Yes, it's slower on a CISC architecture than a RISC architecture --
no big surprise there.  It's a benefit of running on RISC, at the
expense of a few extra cycles for the same task on occasion.


If you're complaining that it only runs on a small subset of x86
hardware, that may be a valid complaint against the x86 port of
Solaris, but it has no relevance to your claim that Solaris on
x86 is "not fully-functional";  it functions just fine, just perhaps
not on the hardware you'd like to run it on.

So, please better define the phrase "not a fully-functional, 
production environment operating system".  Works just fine from where
I sit; i.e., no better or worse than the same configuration on
a Sun-branded box, modulo the expected performance hit from running
on x86.  And if that is what you meant that phrase to mean, perhaps
you should re-think it.  "Slow" does not equate to "broken".  
"Non-optimized", "resource-starved", or "underspec'd", perhaps.
But not "broken".
 


> 
> Despite the fact that BSD may appear to run fine on x86, I'd like to 
> point out that linux does too.  But set up several linux machines, 
> running on x86, PowerPC, and UltraSPARC, and thrash them thoroughly. 
> x86 performance ends up on the bottom of the heap.  I'll tell you flat 
> out that a 1.8 GHz x86 linux database server won't match an old PowerMac 
> 9600 with a G3/500 upgrade in it, also running PPC Debian.  Been there, 
> done it, the old Mac wins hands down.  P-IV's are nothing but hype with 
> a big heat sink.  You can *just* about heat the server room in the 
> winter time with a couple of those machines.  Admittedly, that's a linux 
> comparison, but OS X is looking to the future as a high-performance 
> operating system that is going to need it's own hardware platform to 
> deliver as promised, just like Solaris does.  Compromise on the hardware 
> to reduce aquisition cost, and you'll also compromise on the 
> performance, stability, and general overall reliability of the system. 
> I'm not saying somebody like AMD won't possibly come up with a decent 
> x86 processor in the future, but right now, I'm not seeing it.


Here, you seem to have gone from complaining that Solaris x86 doesn't
work properly (and being vague, at that), to simple (and unjustified)
platform-basing.

I'm fairly sure you had a point to make.  I'm just not sure what it
is.  I am, however, fairly certain there weren't any facts in the
vicinity lending them support.  

I've seen various statistics that each demonstrate Linux, Solaris, Mac
OS X, Windows (various flavors), NetBSD, OpenBSD, and FreeBSD all 
outperforming one another on the same platform.  The one constant
in each of these studies:  They're invariably done by institutions
with a vested interest in the outcome, and are run by people who 
have no clue how to optimize the subjects of the tests for their
respective platforms.  You can't just take, say, a database, 
and throw it one two radically different platforms, install identically,
and NOT expect one to outperform the other.  That's an unfair
test, because software, by its nature, is biased, from the ground
up.  Optimize the subjects of the tests for the environments in
which they're to run, such that each candidate is making full use
of available resources, and *then* run a comparison, and perahps
I'll take note.


...and before you jump on me, accusing me of attacking whatever
platform you perfer to bolster, please take deep breaths, reread
my post, and understand that I did not once elevate or degrade 
a CPU architecture, operating system, or piece of software; I
merely took systematic issue with your claims, your support for them,
and your clarity of presentation, which limits one's ability to
discern and evaluate the first two.

-- 
Mark C. Langston                                    Sr. Unix SysAdmin
mark at bitshift.org                                       mark at seti.org
Systems & Network Admin                                SETI Institute
http://bitshift.org                               http://www.seti.org