[rescue] Catching up - Mach/Lunas, 2000Es, PPC/88k stuff

[Skeezics Boondoggle]

On Fri, 20 Jun 2003, Patrick Giagnocavo wrote:

> I think you are mistaken about this - was not the earliest Mach effort
> by Rick Rashid at CMU in 1982 or so?  Or perhaps that was the
> precursor to Mach.

The precursor to Mach was "Accent", which was the successor to something 
even more rare called "RIG".  I have no info on RIG... but Accent I know 
quite a bit about, since it was developed on the PERQ workstation.  They 
developed the Accent kernel under the CMU SPICE program, during the first 
half of the 1980s.

Accent's kernel, the disk and Ethernet drivers, and the instruction set
for C/FORTRAN/Pascal all fit within a 16K x 48bit writable control store,
with room left over for a Lisp instruction set (dynamically loaded when
the SPICE Lisp environment was booted - btw, the first workstation
implementation of Common Lisp) AND some room left over for user-loadable
microcode.  Think about that:  kernel, drivers, two instruction sets, AND
room left over.  <dig> Now, if only CMU had developed a windowing system
instead of that heap of cruft we got from MIT...</dig> :-)

Anyway, I had left Pittsburgh when PERQ Systems died in 1985; since it was
no longer a viable development platform, the Accent project morphed into
Mach and was ported off the PERQ to a wide variety of platforms - the
Luna, x86 (I think), and a wide variety of DEC iron, including (both VAX
and MIPS architectures?).  Memory is a bit fuzzy.  But I do know that the
first version of Mach had some code in it that had been run through a PERQ
Pascal -> C translator, and even today on NeXTstep/OpenStep/MacOSX
machines things like the "NetMsgServer" survive from the Accent days...

I'd LOVE to get my hands on a Luna 88k.  Those are cool and funky and
rare, and worthy of collecting. :-)  I'd love even more to find another
working PERQ... my machines are over 20 years old now, and spare parts are
impossible to find.


Two other items - I've been on vacation (first on in YEARS) this week and
am way behind on my email:

sun4d 40Mhz->50Mhz conversion, power costs
------------------------------------------
Yes, an SS1000->1000E or SC2000->2000E conversion is definitely possible;  
change out the control board and move all your CPUs and memory to 50Mhz
system boards and you're there.  You can even move the IDPROM and keep
your old MAC addr and hostid.  No problems.  I did a 2000->E conversion in 
1999 (the machine is still running to this day, coming up on its 10 year 
anniversary! :-)

[skeezics at church:~] 1% uptime
  1:22am  up 231 day(s),  5:55,  13 users,  load average: 1.09, 1.08, 1.09

As for power, a lightly loaded 2000 (8 CPUs, 2GB, some disks) shouldn't
draw more than 1KW or so, on up to a maximum rated load of 3.9KW.  
(Diskless, I've measured my machines at around 700W; with _two_ 30-drive
SSAs they pulled around 1.2KW, a touch more when all the drives were busy.
YMMV.  But that shouldn't be too hard on the pocketbook; I mean, out here
in the PNW we've got "cheap" hydro out the wazzoo, allegedly - I still pay
around 8.9 cents a KWH!  :-/

Someone else mentioned that Cray CS6400 is a sun4d - yes & no; it reports
itself either as a 'sun4d6' or a 'cray4d', depending on phase of the moon
or which "Cray Solaris" release you're running.

NeXT, the 88k, the PPC601, and Irony
------------------------------------
The irony of the role of NeXT in the early PPC/"AIM alliance" days was
that because they had their 88k box (the "NRW" - NeXT RISC Workstation)  
almost ready to go - most of the toolchain was done and allegedly the
kernel was booting - they were uniquely positioned to drop in PPC601's
instead when it became clear that the 88k wasn't going to have enough
volume behind it to succeed.  That the 601 was bus compatible was no 
accident; Apple most likely had 88k prototypes too, and expected to make 
the same conversion.

So IBM, Motorola and Apple spend all that money and work to develop the
new chip, and NeXT would have been the first vendor out of the gate with a
working box... Thus, if Mr. Jobs hadn't killed off the hardware division,
he could have beaten Apple to market with a new machine based on "their"
chip - if that ain't irony I dunno what is.  (And, don't forget, NeXT
would have faced virtually none of the transition issues that plagued
Apple's m68k->PPC transition - NeXTstep already could deal with "fat
binaries" and all the cross-compilation tools were there.)

Those NRW prototypes are extremely rare...I had some emails from a former
NeXT guy a long, long time ago who claimed to have one as well as the
"Nitro" card (40Mhz '040 upgrade board) and a C-Cube compression
daughterboard for the NeXTdimension board... damn, NeXT was so freakin'
cool.  Love him or hate him, Steve Jobs certainly had a way to motivate
people into creating some amazingly cool stuff.  Too bad the so-called
"G5" Macs won't be running "real" G5s... they could have shipped 'em
before X-mas last year if he'd stuck by Motorola, and I think having
*both* IBM and Moto producing high-end PPCs could only have been a good
thing.  The Moto G5 specs posted in early 2002 were staggering - over
1,000 SPECint/fp and clock rates up to 2.2Ghz in the labs... Ah, well.  I
hope we see PPC970's on Monday - better late than never, eh?


Back to my "vacation" - tomorrow it's the sawzall vs. all the old crappy
rusting galvanized pipe in this basement!  I love sweating copper.  Some
people relax by lying on a beach.  I do wiring and plumbing projects.  
Hey, when this whole Internet fad blows over, a guy has to have somethin'
to fall back on!  (Of course, when I'm done with the plumbing, I'm adding
another 100 sq ft of computer room.  It's madness, madness I say! :-)

-- Chris