On Wed, May 30, 2001 at 07:37:49PM -0400, Thor Lancelot Simon wrote:
> 
> Also, there's a bit more about Purdue's box in which they replaced the
> SBI in a 780 with a second processor -- they evidently ended up with a
> spare processor because they dropped another 780 off a truck!

I'm including the below which is a Usenet posting from George Goble
that you can also find by scrolling way down the page at the URL I
mentioned before.  It gives some interesting history about Purdue's
homebrew SMP 11/780s and mentions that there were at least a few
dozen in the field, some built by DEC field service!  If you have a
780 and access to the CPU from another, you might get such a beast
built and able to run NetBSD -- ISTR the actual tech report on how
to do this is still available from George's homepage.

Note the mention of Unix running on the 11/782 internally at DEC.  Given
the way the 11/782 worked, I find that a bit surprising -- anyone else
ever see or hear of this?

| From: ghg@edu.purdue.ecn.en (George Goble)
| Date: Tue, 19 Oct 93 22:06:27 GMT
| 
| In article <CEys3s.I91@nas.nasa.gov>, eugene@wilbur.nas.nasa.gov (Eugene N. Miya) writes:
| >
| > George Goble produced two processor VAXen after reading
| > "It is possible to create a multiprocessor by replacing the SBI with a CPU."
| > He did that after they dropped a 780 off a truck.  But it was not a VMS
| > machine.  A paper exists by George from a Usenix conference.
| 
| We did not drop a 780 off a truck, but we picked one up approx 1 year
| after the first dualcpu was up from a surplus  dealer..
| 
| Most of the credit for the Purdue Dualcpu 780 goes to Mike Marsh,
| Around June 1981, he noticed (in the prints) about something
| about clock jumpers which could be removed for "extra cpu" on the SBI
| termination.  He quickly figured out that the SBI terminator could be
| removed, and replaced with a CPU at a different TR level.. The SBI cables
| needed to be special made, with hot and ground reversed for this..
| We made the first set under a microscope, and later got the cable
| supplier to make them up for us.
| 
| Mike and I instrumented an 11/780, then the "ve" machine in the Purdue
| School of Electrical Engineering, room 338.  I figured out, that most of
| the time, about only 1/3 of memory/SBI bus bandwidtth was being used,
| and it might support a 2nd processor..
| 
| There was a Usenix conference near the end of June, 1981, I think it
| was in Austin??  Mike and I gave talks on Unix Kernel changes and
| general problems with DEC Unibus Boxes, and covered the rewiring of
| the grant lines on the DD blocks to get rid of buss reflections, etc..
| 
| At the close of that talk, I spent 30 sec or so, about how we might
| try to build a Dual 780, by removing the SBI terminator and replacing
| it with a cpu.  Over the July 4 weekend, after returning from Usenix,
| we tried that.. It didn't work at first.. no clocks.  It turned out
| that the "extra cpu" jumpers were incorrect in the prints, and that
| termination was incorrect on the six clock lines when these jumpers
| where changed.  Mike next constructed a termination pack of 12 resistors
| to go on the 6 clock lines, and lo and behold it worked..
| 
| I had a crude master-slave version of Unix up in about 3 weeks...
| We had about 10 crashes/day for those 3 weeks.. but it paid off.
| Armando Stettner (then aps@dec.com?) and I worked on getting
| that Unix to run on the 782, same as the Purdue dual 780 but it
| had MA-780 shared memory controllers (each only 1 or 2 MB), so
| one could easily have a 1$ Million+ config just from the mem
| controllers.  I went out to DEC for a few days and helped APS to
| get it running.. it worked on the 782 as well... I dunno what,
| if anything DEC did with it then.
| 
| The 780 dropped off a truck was not dropped by us. Curt Freeland located
| one with a bent frame at some junk dealer.. At the time SBI cables
| were notoriously flakey. Curt went up to the junk dealer, with
| a set of boards, new SBI cables, etc, to run diags on the suspect
| 780.. It actually worked, even with a bent frame, but had lots
| of bad SBI cables..  Curt marked the bad cables "good", and the
| good ones "bad", and the dishonest dealer, then later removed all the marked
| "good" ones (he had no way of testing them) and put them on other systems,
| and put other unknown cables in their place..  He screwed himself..
| and ended up with a whole bunch of bad ones, and we ended up with only
| 1 or two bad!  The 780 cpu dropped of the truck, after a frame
| straightning, became the slave cpu on ef.ecn.purdue.edu which
| still runs today.
| 
| Purdue built 10-15 dual 780's (there are even two still running
| here today e[af].ecn.purdue.edu).  Around 2500 copies of the Tech
| Report were given out by the EE dept, and it even included
| the parts list to build your own dual 780.. I understand that
| around 30ish were built by local DEC FS's on T & M basis for
| various customers..  The original dual 780 was finally turned off
| and cut up for scrap around 3 weeks ago..it had been renamed to
| ec.ecn.purdue.edu.
| 
| This  dualcpu 780 model & kernel code was used by SEL (which became
| Gould), Sequent, Amdahl, and others as a starting point for their
| multiprocessor Unix machines.
| 
| --ghg  (George Goble)
|
| ghg@purdue.edu

-- 
Thor Lancelot Simon	                                      tls@rek.tjls.com
    And now he couldn't remember when this passion had flown, leaving him so
  foolish and bewildered and astray: can any man?
						   William Styron