4 Dec
2015
4 Dec
'15
10:59 a.m.
From: Eric Smith
Sent: Thursday, December 03, 2015 9:57 PM
On Thu, Dec 3, 2015 at 12:29 AM, Pontus Pihlgren
<pontus at update.uu.se> wrote:
[about KL10/KA10/PDP-6 tri-processor
> Wow, that's impressive. How was it done? Was
it done with DEC or was it
> a local "hack"?
Prior to the 1091 and 20xx, all PDP-10 processors used
essentially the same
memory bus, and the memory boxes were multiported. The necessary hardware
configuration might not have been quite as simple as just cabling the three
dissimilar processors to the memory boxes, but it probably wasn't too
terribly complicated.
Getting standard DEC software to run on such a
configuration would have
required quite a bit of work. DEC supported asymmetric multiprocessing on
the KA10 (DECsystem-1055) and KI10 (DECsystem-1077), and possibly on the
KL10 (DECsystem-1088). Symmetric multiprocessing (SMP) wasn't available in
TOPS-10 until some time after the KL10 was available, and for SMP only
multi-KL10 systems were supported. I think SAIL ran the WAITS operating
system, rather than a DEC OS, though WAITS probably started out as a fork
of an early DEC PDP-10 "Monitor". ("Monitor" was the name of the OS
before
it became TOPS-10.)
My understanding is that the SAIL tri-processor
configuration was
asymmetric multiprocessing. (Not just asymmetric in that the CPUs were
different, but also in how I/O devices were configured on them, and which
CPU the operating system mostly ran on.) However, I wasn't there and only
heard about the system second-hand at best.
SAIL did indeed run WAITS, which officially forked from the PDP-10 monitor at
4S72 (Level 4 Monitor, Summer 1972 release), but which began diverging when it
was still the PDP-6 Monitor.[1] It supports asymmetric multiprocessing, based
initially on the 1055 code[2], though diverging immediately because of the
differences between a KA-10 and a Model 166 processor (PDP-6 CPU).
Things get more complicated with the introduction of the KL-10 processor.
Prior to this, SAIL used non-DEC disks and their own file system, similar but
not identical to the DEC Level D[3]; with the introduction of the Massbus,
they moved to RP06 and RP07 but kept the SAIL file system. However, they did
not adopt the Tops-10 drivers for the Massbus; instead, they modified the
TOPS-20 drivers (as of release 5.1) to interoperate with a Tops-10 style
system call regimen. At the same time, they made the KL-10 the master in the
three-processor system.
Because they started with the PDP-6 and continued development until ~1990
(including porting to the Foonly F2 at CCRMA and the KL-10 at Livermore),
there was no such thing as a WAITS install tape (or suite). That made getting
WAITS running on a system at the museum a long and winding road, an adventure
in dissertation level research, and that in turn is why I know so much about
the internals and history of the operating system.
Rich
[1] "Tops-10" was simply a renaming of an operating system which began on the
PDP-6 in 1964 and continued in an uninterrupted line of development up
through the final release, Tops-10 v7.04 (1988), and maintenance (v7.05,
1993).
[2] The PDP-10 monitor, when introduced, came in 5 variants, of which 2 seem
to have been vaporware (to use an anachronistic term). These went from a
single-user monitor (10/10) up to a timeshared swapping monitor with disks
(10/50). This was the version to which AMP ("10/55" " was introduced,
and
really the only version supported beyond initial release.
[3] For example, project-programmer numbers are SIXBIT rather than numeric,
even when they look numeric. In a DEC file system, the Master File
Directory is [1,1] and in the system that is represented by the octal
value 000001000001; in the WAITS file system, the MFD is [1,1] with an
internal representation of 000021000021.
Rich Alderson
Vintage Computing Sr. Systems Engineer
Living Computer Museum
2245 1st Avenue S
Seattle, WA 98134
mailto:RichA at LivingComputerMuseum.org
http://www.LivingComputerMuseum.org/