[rescue] 3b1/7300/unixpc/s4 history [was: Wanted - 3B2]

Geoffrey S. Mendelson gsm at mendelson.com
Wed Jul 9 09:04:17 CDT 2008

On Wed, Jul 09, 2008 at 09:42:24AM -0400, Sandwich Maker wrote:

> the sV kernel worked fine, but it was designed to be portable and not
> play to the strengths or weaknesses of any particular architecture.
> the drivers would be a reflection of that.  iirc this was the crap the
> hp developer referred to, that nothing was tuned to hp-pa system
> architecture.

I can understand that, but wasn't it HP's job to tune it. That's why AT&T
did not, and sold source licenses. It's actually a good idea commericaly.
Sell "UNIX" (they could sell binaries, but not use the name) which 
did not perform well, and "sell up" to improved versions. Since you already
had the binary license for that version ($60 for 1/2 user(s), $250 
for unlimited users), all you had to pay for was "support" and "improvments",
which would be far cheaper than another version or a competing product (with
a new license fee).

I would love to know how it would have performed on the HP3000 architecture,
which as real stack processor. C seems well suited for it.

> the att std c compiler was likewise also designed to be portable.  the
> unix team was actually proud of their lowest-common-denominator
> portability approach, and how rapidly they could port sV to a new
> platform as a result.

That was the guiding light behind GCC too. It came about because Sun included
a C compiler with SunOS as you needed it to maintain the kernel. System V 
(usually misnamed Solaris) did not need a compiler, some of the kernel was
"built on the fly", other parts just needed the linker, so Sun did not
include a C compiler.

> iirc dec unix got about twice the mips from a vax that att sV did,
> because of all this.  it's not a complete apples-to-apples comparo b/c
> dec unix was of course berkeley based.

There is a big difference between BSD in those days and SysV, I expect now
that the remaining BSD variants support as much as SysV does, they would
not be much faster if at all.

> this could partly be because rewriting drivers is much easier than
> rewriting the kernel, and you have to write new drivers anyway for
> new hardware.

True, but the AT&T drivers were ones that were the simplest to write.
I once read a book about writing UNIX device drivers for the PC, but the
closest I got was writing them for Coherent, if you remember it.

> had sco been spun out of microsoft by that time?  they were originally
> formed to supply unix for the 68k-based tandy pc.

Yes, they were selling a SysV UNIX at that time. I can't remember if it
was under their brand (Xenix?) or a different one. Wasn't the Tandy one 
sold under Xenix?


Geoffrey S. Mendelson, Jerusalem, Israel gsm at mendelson.com  N3OWJ/4X1GM

More information about the rescue mailing list