[Info-vax] Y3K for PDP-11 Operating Systems

[Johnny Billquist]

On 2011-05-06 12.08, G Cornelius wrote:
> Richard B. Gilbert wrote:
>> I have no PDP-11.  I had one at work many long years ago.  I don't miss it.
>
> You must never have worked on an -8!

Heh. Yeah, well, the PDP-8 code is pretty compact in a way, but of 
course you don't write something like SSL for a PDP-8.
Also, on both the PDP-8 and PDP-11 (but especially on the -8) assembly 
language programming was the norm. And you get so much more use out of 
memory that way. (And it also encourages you to keep your programs 
small... ;-) )

> A 22-bit PDP-11 with RSX-11M+ (I/D space, supervisor mode libraries)
> was a huge improvement over the sad little 18-bit beasts (physical
> address, of course). And memory resident overlays at least meant
> that you did not have to wait for more than a single system call's
> overhead for additional segments of your overlaid program to be
> accessed. But, oh, the task builder overlay description language
> files you had to manage!

Yeah. The ODL-files are not that fun, but I agree that M+ on a big 
PDP-11 is a very different experience to some small -11 with some other OS.

> The 18-bit physical address space was a broken concept from the
> beginning: "Hey, we have all this 18/36 bit stuff already built
> for the 10- and 20-series, let's give the 11's two entire bits
> of extended physical address!"  To go to 22 bits your device
> drivers still had to go through the 18 bit atrocity, with the
> extra two bits tucked into the CSR somewhere, just so they could
> address a set of "Unibus mapping registers" to extend the map
> from 18 bits to 22 bits. In retrospect it all seems to have been
> rather poor planning.

Well. The 18-bit thingy was not totally for the PDP-11. The Unibus was 
first done for the PDP-11, true, but the PDP-11 at that time was only 16 
bits (there was no MMU). However, the Unibus was also used in some 
18/36-bit products, where those two extra bits made sense. And sharing a 
bus design between several machines also makes sense (and both address 
and data can be 18 bits).

But yes, having room only for 18-bit addresses on the bus was obviously 
pretty little, and the PDP-11 quickly grew out of that too. Gordon Bell 
have a famous quote for the PDP-11 and address extensions.

The Unibus map is annoying, true, but not really a big issue. It's the 
same thing on a VAX. Except the Unibus map is more fine grained on VAXen.

(That's why the 11/70 had Massbuses... ;-) )

> If in fact your code fit into the address space and did not need
> 32 bit arithmetic, it could be fast.  The first time I started
> working in a group that had VAXen I noticed two 750's sitting
> off to the side because they had not been able to handle the
> application load.  Their replacement: a pair of 11/84's.
> Granted, this was specialized in that the language interpreter
> had to do a lot of single byte operations, resulting in the
> initial versions, on 750's at least, being rather inefficient.
> Sometimes extra bits just mean extra overhead.

Heck. The 11/750 is not a fast machine. It's definitely slower than an 
11/84, unless you do things that the VAX can do natively and the PDP-11 
can't.

	Johnny