Future of Unix - comments please
For several years a number of us have tried to convince a company
called MathSoft to port its S-PLUS statistical computing environment
to Linux (or at least to some Intel-based variant of Unix). They keep
putting this off because they feel that Linux is not standardized or
not "industrial-strength" or ...
Of course, I feel they are missing the boat badly here. There is now
a freely-available version of the language that lives very happily on
Linux. (Yes, I will make a package of it. I'm waiting for the new
upstream version due out RSN.)
However, I still have a need for the commercial system so I can ensure
that my code will work with it. I want to convince the company that
GNU and GNU/Linux are viable operating systems worthy of consideration.
I would even go so far as to say that the only operating systems with
long term viability are MicroSoft's products and versions of GNU such
as GNU/Linux. I outline some of my arguments on this below. I would
appreciate comments and corrections.
1. In the near future Intel will dominate the microprocessor market even more strongly than it does now.
Bruce Perens recently wrote that SGI is abandoning the MIPS
architecture and switching to Intel microprocessors. I didn't know
that but it doesn't surprise me. Consider which architectures besides
Intel have a substantial part of the Unix market presently.
MIPS - See above.
HP-PA - Hewlett Packard is co-developing Merced with Intel.
Alpha - DEC is suing Intel, Intel is counter-suing DEC. One recent
report suggested that a possible resolution is for Intel
to buy the Alpha architecture from DEC. Even if that
doesn't happen, Alpha may disappear. It is elegant and
powerful but cannot seem to get the market momentum.
PowerPC - Apple is on the ropes and both IBM and Motorola are having
serious doubts about PowerPC.
SPARC - Still viable for a while. Until people do price/performance
comparisons they often believe that UltraSPARC's are very
powerful. When you actually benchmark something like an
UltraSPARC against a Pentium Pro-based system you often
discover that the Pentium system is considerably faster yet
costs just a fraction of a comparable SPARC system. Someday
the Sun user base will discover this.
I don't think I have missed any major systems there. Please tell me
if I have.
It looks like there will not be future versions of HP-PA and MIPS. I
think PowerPC and Alpha are both pretty shaky right now and SPARC is
going to run into trouble. Are any of these going to be able to
compete with Merced? The figure I have heard is that the design of
Merced will cost about $100,000,000. Does Sun or DEC have comparable
amounts of money to invest in the _design_ of a processor not to
mention the cost of building the fabrication facilities?
On the long-term viability of SPARC - I have several colleagues who
have purchased UltraSPARC 2 systems, usually for $10K to $20K. They
really feel foolish when they find out that my dual Pentium Pro system
for about $3500 outperforms their systems. A group I work with was
going to purchase a quad-processor memory-rich compute server. The
quote from Sun was around $100K. A comparable Intel-based system with
Linux pre-installed was around $11K. They went with Linux.
2. If most manufacturers are using Intel processors, then the market
is essentially an open hardware market.
There will continue to be some misguided attempts to close hardware
standards but they will not succeed. The market has gotten too
accustomed to having open hardware standards to go back.
3. The open hardware boxes will be shipped with some variant of a
MicroSoft operating system installed. Vendors of other OS's will have
to convince users why they want to replace that operating system.
That is not an easy sell unless you are talking about something like
GNU where you can show superior quality at little or no additional
cost.
4. Open software is no longer playing catch-up with commerical
software. It is leading.
I sent a message about the work by Bart Miller and colleagues on
trying to crash Unix utilities by sending random input streams at
them. The GNU utilities and GNU/Linux as an operating system were the
_best_ of all the systems tested - far better than the commercial
systems. See
ftp://grilled.cs.wisc.edu/technical_papers/fuzz-revisited.ps.Z
Code is in
ftp://grilled.cs.wisc.edu/fuzz/
This is just one example but we know it is a common occurence. The
freely available software ends up superior to the commercial version.
RMS was right.
We all know the arguments for this. If you ship the source code then
others can discover and correct the errors, etc.
Regretably there are still many managers and administrators who simply
cannot conceive of the freely available system being superior to the
one for which you are paying money. I would appreciate any
observations or examples you can provide for me to help in my attempts
to convince these managers that Linux is for real.
--
Douglas Bates bates@stat.wisc.edu
Statistics Department 608/262-2598
University of Wisconsin - Madison http://www.stat.wisc.edu/~bates/
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-devel-request@lists.debian.org .
Trouble? e-mail to templin@bucknell.edu .
Reply to: