4.2BSD on uVaxII ??

[Pete Kaiser, HLO2-1/N10 225-5441]

Sebastian Schmitz writes about the MicroVAX II, and has some questions and
comments.  I hope I won't be the only one to reply to them, and in particular I
hope some of the MicroVAX II field test sites, and especially the ones that have
been running Ultrix-32m, will also respond.  Let me also say at the outset that
I admit that (a) I work for Digital, and (b) I think the MicroVAX II is a very
nifty engine.

> The new MicroVAX ][ is the latest VAX: its a single chip VAX
> (so they say) and its claimed to have 80% of the power of a
> 780. Its meant to support up to 48 users but typically it will
> be less.

It really is a single-chip VAX.  The chip is a double-metal NMOS chip with about
125000 transistor sites and on-board memory management.  Depending on applica-
tion, the proportion of a VAX-11/780 it represents may be up to more than 100%;
however, around 80 - 90% is a more typical range.  And I wouldn't, personally,
say that it's "meant to support up to 48 users", although you could certainly
hook that many up to it.  For years lots of people have said, in one phrasing or
another, "Wouldn't it be great to have a 780 all to myself?"  And that's what
it's for: to be a very powerful, affordable 780-class machine for a small number
of users.

> Price wise it seems competitive: an MV ][ with a dual floppy
> amd integral streamer cassette tape (95 MB), 2MB Ram, 71 MB
> winchester and Ethernet interface will cost less than DM 100k.
> (For B-movie actor fans this means about US$ 30k)

The key is "affordable".  The lowball system, which includes 2MB of memory,
disk, floppy drive, console port and Ethernet controller, is less than $19K in
quantities of 1 domestically.  The point is that it shouldn't be necessary to
spread the cost among 48 (or even 30) users to make it economical.

> The backplane will support two controllers for discs but the
> floppy drive and tape drive are included so you can presently
> have at most 3 * RD 53 (@ 71MB each = 121 MB) in there with everything
> else. Why on earth will two controllers only support five
> discs ? 2.5 discs/controller doesn't look right somehow. In
> other words I don't know.

Sebastian writes in his note about what are several different standard system
configurations.  In this paragraph he has in mind one of the ones in the BA123
cabinet, the "World Box".  Leaving aside what 3 * 71MB equals, I'll correct him
slightly as regards the U.S. domestic market: currently up to four 5-1/4-inch
form factor devices may be mounted in the BA123.  I believe the standard high-
ball configuration in the BA123 includes three RD53 Winchester disks plus a TK50
(95MB streaming cartridge tape).  There are actually five slots for devices in
the BA123, but although the power supply will drive five in steady-state opera-
tion, it won't quite stand the current surge at startup, so for now only four
are allowed.

> The actual architecture: its a full VAX but of course no PDP11
> compatibility mode. Its a Q-Bus machine and the memory is on a
> special 32bit bus. This is different to the MV I which had a
> dual ported memory used by the CPU and the I/O stuff via the
> Q-Bus. This was one of the reasons why the MV I was/is so
> dreadfully slow. The MV ][ cures this with the special bus (but
> I wonder what happens when those dreaded DMA devices come
> along).

The MicroVAX I doesn't have a dual-ported memory; the memory is on the Q-bus,
pure and simple.  This means that the CPU and I/O devices must contend for the
bus, which does indeed slow down the CPU.  The MicroVAX II's memory is off the
Q-bus entirely, but mapped to it through scatter-gather registers just as the
Unibus is mapped on macroVAXes.  This has some interesting repercussions.  The
first is "what happens when those dreaded DMA devices come along": nothing.  The
CPU can access memory at absolute full speed (6MB/sec) at the same time the Q-
bus is accessing memory at its full speed in DMA "bus hog" mode (3.3MB/sec),
because the memory's speed is 10MB/second.  In other words, the machine can
compute and do I/O both at full speed simultaneously.

The second repercussion of the memory's interaction with the Q-bus through the
scatter-gather map is that Unibus drivers have been copied off 780s and just
dropped onto the MicroVAX II ... and worked.  On Ultrix-32m.

> The $64,000 question: what about UNIX. Well needless to say not
> 4.2 (due to lack of Q-Bus device drivers) but rather ULTRIX 32m
> (for MicroVAX).

I've said it here earlier, and I'll say it again: Ultrix-32 is 4.2BSD.  The one
concession to the MicroVAX is that a rather small list of things is omitted from
Ultrix-32m (the MicroVAX release of Ultrix) to avoid using valuable disk space.
I'm sorry to say I don't remember the entire list, even though I have Ultrix-
32m, but it's nothing that bothers me much.  For instance, I can live without
on-line documentation, since I have it nearby on a shelf anyway.  And it does
seem reasonable that a Q-bus machine needs no Unibus drivers.

> [W]hy would people want such [a] machine?

Left as an exercise for the reader.

> I feel the only reason to get such a machine is when
> you want to write assembly language progs which are binary
> portable within the VAX range. Ahh but then are they ? How
> about system calls,etc.

They are.  You can take your executables from any VAX, copy them onto a MicroVAX
II and run them as is.  There's one simple exception to this rule: hardware-
dependent code.  (I trust this is no great surprise, since the same rule holds
true even among macroVAXes.)  Hardware-dependent code includes references to
privileged registers and some things about I/O -- although, as I mentioned above
about drivers, some surprising things work identically between macroVAXes and
the MicroVAX II.

I hope I've stuck sufficiently to the facts here.  Questions?  Flames?

---Pete

Kaiser%JAWS.DEC at decwrl.arpa, Kaiser%BELKER.DEC at decwrl.arpa
{allegra|decvax|ihnp4|ucbvax}!decwrl!dec-rhea!dec-jaws!kaiser
DEC, 77 Reed Road (HLO2-1/N10), Hudson MA 01749		617/568-5441