[TUHS] UNIBUS issues (Was: Spacewar at Bell Labs)

Noel Chiappa jnc at mercury.lcs.mit.edu
Sun Feb 9 04:59:42 AEST 2020


   > From: Dave Horsfall <dave at horsfall.org>

    > [ Getting into COFF territory, I think ]

I'm sending this reply to TUHS since the message I'm replying to has some
errors, and I'd like for the corrections to be in the record close by.

    > On Thu, 30 Jan 2020, Clem Cole wrote:

    >> They way they tried to control it was to license the bus interface chips
    >> (made privately by Western Digital for them IIRC but were not available
    >> on the open market).

Although DEC did have some custom chips for QBUS interfacing, they didn't
always use them (below). And for the UNIBUS, the chips were always, AFAIK,
open market (and the earliest ones may have predated the UNIBUS).

E.g. the M105 Address Selector, a single-width FLIP CHIP, used in the earliest
PDP-11's when devices such as the RK11-C, RP11 and TM11 were made out of a
mass of small FLIP CHIPS, used SP380A's for its bus receivers and 8881's for
transmitters.

On the QBUS, the KDF11-A and KDJ11-A CPU cards used AMD 2908's as bus
transceivers, even though DEC had its own custom chips. The KDF11-A also
used DS8640's and DS8641's (transmitters and receivers), and also an 8881!
(The UNIBUS and QBUS were effectively identical at the analog level, which is
why a chip that historical was still in use.)


    >> If I recall it was the analog characteristics that were tricky with
    >> something like the BUS acquisition for DMA and Memory timing, but I
    >> admit I've forgotten the details.

One _possibility_ for what he was talking about was that it took DEC a while
to get a race/metastability issue with daisy-chained bus grant lines under
control. (The issue is explained in some detail here:

  https://gunkies.org/wiki/Bus_Arbitration_on_the_Unibus_and_QBUS

and linked pages.) This can been seen in the myriad of etch revisions for the
M782 and related 'bus grant' FLIP CHIPs:

  https://gunkies.org/wiki/M782_Interrupt_Control

By comparison, the M105 only had 3 through it's whole life!

It wasn't until the M7821 etch D revision, which came out in 1977, almost a
decade after the first PDP-11's appeared, that they seemed to have absorbed
that the only 'solution' to the race/metastability issue involved adding
delays.

In all fairness, the entire field didn't really appreciate the metastability
issue until the LINC guys at WUSTL did a big investigation of it, and then
started a big campaign to educate everyone about it - it wasn't DEC being
particularly clueless.


    > Hey, if the DEC marketoids didn't want 3rd-party UNIBUS implementations
    > then why was it published?

Well, exactly - but it's useful to remember the differening situation for DEC
from 1970 (first PDP-11's) and later.

In 1970 DEC was mostly selling to scientists/engineers, who wanted to hook up
to some lab equipment they'd built, and OEM's, who often wanted to use a mini
to control some value-added gear of their own devising. An open bus was really
necessary for those markets. Which is why the 1970 PDP-11/20 manual goes into
a lot of detail on how to interface to the PDP-11's UNIBUS.

Later, of course, they were in a different business model.

       Noel


More information about the TUHS mailing list