On Wed, 20 Jan 1999, Douglas W. St.Clair wrote:
> >On Fri, 15 Jan 1999, Herbert Wengatz 42850 wrote:
> >
> >> At that time Sun's intention was that an OS is only the vehicle
> >> for selling hardware. Not a product itself, since they didn't
> >> wanted to make themselves concurrence.
>
> To which Robert Brown provided a long and thoughtful reply. The essence of
> which
> seemed to me to beg the question is Apple, Sun, etc. in the hardware or
> software business? I should like to suggest that, without questioning his
> assertions, the following pattern has been followed. Build simple hardware
> and provide functionality with software. As more complex hardware becomes
> possible embed "useful" functions in the hardware because hardware is
> faster than software. Now if you accept this as a viable model you may
> believe you need to provide both hardware and software as a package as
> Digital, IBM, Sun, SGI, etc. have done. It isn't that Sun or Apple is
> inherently confused about whether they are a hardware or software company
> it's just that they need to control both if they accept the model I
> suggested above. Now how would you get them to abandon this model and what
> model should they adopt is a question that interests me.
> END
> **************************************************
Except that for years Apple's systems ran on commodity processors like
the Motorola 68K, and it still runs on commodity processors like the
Power PC (where I define a commodity processor as one that sells to more
than one system manufacturer and runs (or can run) more than one
operating system). And frankly, how could one ever tell a "fast" apple
from a "slow" apple? The GUI that has been the centerpiece of Apple's
efforts forever now is immensely slow (as the sage says, "Learn to use a
Macintosh in a day, pay for that knowledge for a lifetime") and totally
nontaxing for any modern CPU. But since Apples for most of their
lifetime have lacked anything like a coherent, open networking
interface, a proper networking security layer, a shell (so one could do
something over the network with an apple one is not sitting at) all one
ever does with an apple is personal computing. Gee. Watch the wasted
cycles build up even faster...or display those cute little screen savers
that burn cycles making pretty, mindless, pictures.
Sun for many years (a decade or so) has claimed that SPARC is the
technically superior RISC implementation (and it does indeed have a lot
of nice features) but its CPU performance has historically over that
period been no better than average. I honestly think that they (and
possibly Intel, in their latest move) have started to move CISC-y
garbage like specialized graphics processing or decompression commands
back onto their CPUs because they have been unable to produce silicon
that competes well in raw integer/float price/performance with the
higher clock Alpha (and sensibly recognize that the market of folks that
really care about floating point performance is far smaller than the
market of game/graphics freaks). That is, I think it is a strategy
designed to minimize an ongoing failure, not a plan for success.
Digital's Alpha is also a poor example for the argument you seem to be
making. It is arguably the world's best commodity CPU (one of the best
price performers as well as the best raw performer, especially in
floating point) and is by far the "RISC"iest of the RISC CPUs. Its
strength is its simplicity, which lets Digital get a lot of mileage from
expensive silicon real estate at the standard tradeoff of making
compilers work a lot harder to optimize code off-chip. I have yet to
see a SPARC (or any other) chip that beats the alpha's SPECmarks at
equivalent single CPU system cost, or that beats the clock speeds that
the alpha can achieve, and clock speed alone is the dominant system
performance indicator. IMHO, or to my humble experience, anyway.
I wouldn't wish IBM's AIX distribution on a dog 'cuz I like dogs,
regardless of the platform it ran on. Quite literally, you couldn't
give me an AIX cluster if I had to run AIX on it. So I cannot honestly
address whether or not IBM is adding more complex hardware with their
own secret software hooks for it and hope I never have to find out. If
so, this too is a strategy of failure. The world loves COTS these days
and is very cruel, in the long run, to specialists, who at best carve
out and narrowly maintain, for a time, niche markets.
The one system manufacturer that I think supports your model well is
SGI. SGI has dabbled in the "desktop" workstation market with offerings
like Indigo, but for the most part has remained solidly centered in
offering expensive, high margin systems with features that a small class
of users consider indispensible. Obviously, in the early days this was
graphics, and their integrated frame buffers and software were
world-class; they also have maintained a long-standing presence in the
low end supercomputer business. Nowadays, their software is still world
class, but COTS pressure has put the serious hurt on the frame buffer
business with every TD&H making high end video adapters for gamesters at
ever-lower margins. I think that one could argue that some of the best
video adapters available today run on the PCI bus (or AGP bus) in PC's
and are remarkably cheap. Even their graphics software advantage is a
bit cramped as PD clones get ever better.
Nevertheless, they do manage to come up with new, aggressive hardware
mods that maintain a REAL (not failure-driven) competitive edge and that
yes, generally need their software to exploit. They basically own their
own MIPS CPU that has its own special virtues, although thus far it too
hasn't managed to touch the Alpha on price/performance. But aren't they
one of the main forces behind CC-NUMA, for example? So I'll give you
SGI...
Still, linux runs on nearly all of the CPUs or systems produced by the
manufacturers listed above and from what I've read on this list (not
from personal experience) often outperforms the manufacturer's "own" OS
on those CPUs/systems. I would therefore claim that a close examination
of all the small computer manufacturers of the last decade to a decade
and a half strongly supports the contention that the manufacturers of
hardware are pretty clueless when it comes to planning a marketing
strategy. Over many years we can see that they:
a) Attempt to tie their software to "their" hardware only. Curiously,
they do this even when their software is notoriously worse than
everybody else's and their hardware might just be better. See "Data
General and the M88K". They also do it when their software is better
(in at least some ways) than everybody else's but their hardware sucks
in price/performance. See "Next and the M68K".
b) Often, they actively or passively oppose the porting of other OS
software to their hardware. As it "You ain't runnin' no Windoze on this
Mac, chum.". Or the porting of their software to other hardware.
Apple's OS on Intel? Scandalous! Never! Of course if they had at the
right time APPLE would be Microsoft (whether or not their hardware
business survived).
c) Have totally failed to see the hardware COTS revolution even as it
happened. This one really has me puzzled. Can it be that (fill in one
of Sun/Apple/IBM/SGI/Next...) don't really WANT to suffer what Intel has
suffered as it has sold system after system to every corporation and
most households in America and much of the rest of the world? Are these
people brain dead? Do they imagine that they will be able to continue
selling high margin hardware forever in the face of Moore's Law and
relentless consumerism? They have one and all put themselves in a
position where they are doomed.
d) Have totally failed to see the software COTS revolution even as IT
happened. When did you ever see a Sun Sparc application in a bookstore?
Or Babbages? Or Best Buy? Or Wal Mart? Sure, Sun makes a lot of money
for each application that they sell at huge prices for their hideously
expensive hardware, but they only sell a handful of applications because
there aren't that many fools (or people who really need and want them)
in the world. Microsoft, on the other hand, probably makes money every
time you sneeze, because their software is relatively cheap and hence is
ubiquitous. A mere hundred dollars of profit apiece, per year, from
twenty or thirty or fifty million people is a fairly hefty chunk of
change.
When I look into my crystal ball, I see a world five years from now that
is much like the current world with a few minor changes. Computers will
be ubiquitous, dirt cheap, amazingly powerful by today's standards, and
will all be connected to the Internet by fairly powerful (1 Mbps or
better) uplinks (all gimmes as these aren't very insightful predictions,
really). Nearly all of these systems will be running linux or freebsd
or perhaps a new (but free/PD) OS that goes beyond either one. They
will not, for the most part, be running Microsoft anything, as Microsoft
will be in serious financial difficulty as they eke out a slowly
shrinking existence on legacy stuff. By this time, only complete idiots
will pay more than $50 for their OS distribution, which will come
complete with a high quality, free, web tool (browser will be way too
modest a term) that is integrated with a free office suite (word
processor, spreadsheet, real (SQL) database interface, email,
multimedia, and more) and a bunch of free games. IRC will have been
replaced by an ongoing, globe-spanning, anyone-can-join-or-leave
networked gameset that will be free, or almost. Virtual worlds...
Actually, there will be damn little commercial utility software sold to
consumers off the counter, because linux users don't need most of it now
and its not like the tools they have now will go away. Rather they will
be strongly and continuously improved and augmented. Prices for what IS
sold will be very, very low, anticipating large volumes at modest
margins or bust. The GUI outside of the web tool will still exist and
will interface either directly or via the web tool to launch or connect
to games. And there will still be a command line/shell interface! Game
software will be one of the few areas where programmers can still make a
commercial killing, but it will be a whole lot more like writing a novel
than inventing a machine. Network/distributed applications (games,
databases, more) will be ubiquitous as protocols are developed for
constructing them and securing them transparently to the user, whose
sophistication will be greater but still not all that great.
And the computers? Alas, I see fewer and fewer to choose from, even as
there are more and more to choose from. This continues the trend of the
last five to ten years. I don't see ALL the hardware architectures
going away -- survivors will be Intel and all its cloned children (some
of whom will by then really have surpassed the parent), the Power PC
(maybe -- if it survives a possible early abandonment by one or two of
its major users well), the Alpha (got a soft spot in my heart for the
Alpha). MIPS will go bye bye. So will Sparc (sorry, Sun). IBM will
continue to use its own chip(s) as long as it pleases -- they already
have a huge base of very stupid customers and I see no reason for this
to change. In fact, as stupid customers finally abandon Sparc and SGI
abandons MIPS to take a step closer to being a COTS-plus company, their
market of stupid prospects could actually increase.
I'd rate the probability of introduction of a totally new concept CPU
over this interval at close to unity, and the probability of survival of
the new concept CPU(s) at close to thirty percent. To survive, though,
there has to be a large enough niche market to permit the product to
survive to COTS maturity and the consequent hyperinflation of market and
hyperdeflation of price. I think that is what SGI is banking on (and
they may yet make it). I think that Sun thinks that the server market
will "save" them -- the fools! My current desktop could be a "server"
for a rather large fraction of the corporate market at a couple of $K
plus disk and backup a la carte. I could build a whole 100 GB SMP
server cluster for what Sun charges list for one not-too-impressive
workstation. I have no idea what Digital/Compaq are planning, but I
have a lot of respect for Digital's engineers -- they may have something
very cool on tap. Some of the possible new concept CPUs might involve
new physics (or new engineering of old physics), though, which makes
them very difficult to predict.
Even without a new concept CPU, though, pure Moore's Law scaling
predicts some 10-50 times more raw power per CPU, with $1000 systems
running 10's of billions of instructions (float or integer) per second.
A local hard disk will be capable of holding approximately a terabyte.
These systems will come equipped with 32 Gbytes of memory standard
(expensive ones with still more, of course). Sure, they'll recognize
voices and the like -- badly -- but we'll still be half a decade short
of systems that are really intelligent.
Maybe these will all end up wrong, but it will be fun finding out either
way...
rgb
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:[EMAIL PROTECTED]
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
