[Info-vax] DEC Large Systems

[Rich Alderson]

=?UTF-8?Q?Arne_Vajh=c3=b8j?= <arne at vajhoej.dk> writes:

> On 6/27/2019 7:50 AM, Bill Gunshannon wrote:
>> Another question born out of curiosity.

>> I was looking at a catalog called Large Systems Software
>> Referral Catalog (Fourth Edition) from 1984.

>> It lists DECsystem-10, DECSYSTEM-20 and VAX. (notice the
>> curious spellings used by DEC themselves  :-)

>> What exactly did the VAX offer or what shortcomings were
>> seen in the 10 and 20 that caused VAX to become the hero
>> and the other large systems to go away?

> This is before my time.

> But my understanding is that the old boxes had a memory
> limit of 4 MW while VAX had larger byte address space
> (32 bit virtual and 24/30/32 physical).

Addressing on the PDP-10 line grew with each model.

The original PDP-10 (KA-10 processor) addressed 18 bits (256K) of
36 bit words.

The second generation DECsystem-10 (KI-10 processor) had a simple hardware
pager which addressed 22 bits (4MW).  Processes were limited to an 18-bit
address space.

The third generation DECsystem-10/DECSYSTEM-20 (KL-10 processor) had the same
simple hardware pager as the KI-10, and in "Tops-10 paging mode" was limited to
22 bits with 18-bit process spaces.  However, TOPS-20 microcode implemented a
pager based on the BBN hardware pager for the KA-10 (!) which implemented 23
bits of virtual address.  Processes were still restricted to 18 bit addresses.

TOPS-20 v4.0 changed all that.  The microcode pager was extended to 30-bit
addressing, although the hardware on the KL-10 was still limited to 4MW
physical.  The monitor ("kernel") utilized "extended addressing" to move beyond
single-section (256KW) forks.

TOPS-20 v5.0 allowed *user* programs to go beyond single-section addressing.

Tops-10 went to "Tops-20 paging" in v7.01, with user programs allowed to use
extended addressing in v7.04 (1986).

"Jupiter" was supposed to be faster than the KL-10 by a factor of 3 or so; that
was far too slow given how long it took to get to that point, so the
DECSYSTEM-40 was canceled in 1983.  The uproar from the installed customer base
was so great that Digital was forced to continue hardware development for
another 5 years (giving us the CI, HSC-50s and RA-81 disks) on the KL-10, and
software development and support for another 10 for both Tops-10 and TOPS-20.

> So DEC either had to make significant changes to the
> old architecture or go with a new architecture.

They tried to make significant changes to the old architecture, and blew it.

> We know what they chose. Could they have chosen different: likely.
> That is like asking if VAX could have been extended to 64 bit. It
> probably could, but DEC chose to go with Alpha.

By way of comparison, the XKL Toad-1 System is an extended KL-10 clone which
implements the 30-bit extended address in hardware, with (1995) modern
peripherals, a better Ethernet interface than the NI (one based on the Stanford
University experimental Massbus-based interface for the KL-10), and a much
lower cost.

Digital could have done it.  They were in the thrall of the 32-bit world by
that time, with Gordon choking his own baby.

-- 
Rich Alderson					  news at alderson.users.panix.com
      Audendum est, et veritas investiganda; quam etiamsi non assequamur,
	  omnino tamen proprius, quam nunc sumus, ad eam perveniemus.
									--Galen