From: Michael Kerpan
Sent: Thursday, January 15, 2009 9:20 AM
There's recently been a lot of talk about the XKL
TOAD, and I, as an
interested lurker had been holding my tounge, despite having some
questions. No more.
1. How does the XKL compare with the Jupiter? Given
that they both had
features like an extended address space, was the XKL basically a "if
DEC won't build it, we will" type of thing?
The Jupiter (KC-10 processor) was physically very large, and designed to
use the next generation(s) of DEC peripherals. It was implemented using
technology similar to the KL-10, with a CI bus to HSC-50s. It was
supposed to be a lot faster, but failed to meet specs.
The Toad-1 System was the result of Len Bosack's suggestion in the mid
1970s (as the KL-10 was going out the door) that DEC should build "a 10
On A Desk." It was not originally intended to be a server. (This was
the original product plan of cisco Systems, by the way. TCP/IP routers
were simply a cash cow to generate revenue for the development effort.
The VC Board of Directors saw things differently when the money rolled
As finally built, it's the size of a two-drawer file cabinet. The XKL-1
CPU uses a couple of Altera FPGAs and a 2916 microsequencer as the
microcode engine. Depending on instruction load, it is 1.8 to 2.5 times
as fast as a KL-10.
2. What the difference between the XKL TOAD and the
systems? Was SC a reseller/packager/intergrator or did they have their
own seperate "Super KL"?
Systems Concepts started in the 1960s. Their first product, the SA-10,
was a KA-10 I/O bus to IBM 360 bus-and-tag protocol converter, which
allowed PDP-10 systems to use IBM disks (and tapes, I believe).
Mike Leavitt and Stew Nelson, along with Pete Sampson, Fred Wright, Stu
Grossman, and a couple of others whose names I've forgotten 20 years
later, designed and built the SC-30M (code name Mars--as Fred put it in
their marketing brochure, "Mars is a lot smaller than Jupiter, but it's
a lot closer.") to be a KL-10 clone that ran faster. It was implemented
in TTL instead of the KL-10's ECL, and was pipelined (something people--
even at XKL--think you can't do with the PDP-10 ISA) for greater
throughput. It originally came with SA (bus-and-tag), CI, and EI
(NIA-20 compatible Ethernet) interfaces; they got the MI (MASSBUS)
working at our (Stanford LOTS) insistence. It maxed out at 8MW physical
memory, but the work to take TOPS-20 from 4MW to 8MW was lost in a disk
crash on their KI-10. It was about 2.2 to 2.4 times as fast as a KL-10.
The SC-30M was designed as a multiprocessor system, to support Tops-10
SMP in a single box. They repackaged a uniprocessor version as the
SC-25 (and as the Sc-20, with a half-speed clock). The big follow-on
was the SC-40, with floating-point hardware that pushed the performance
of FP operations to more than 10x of the KL-10; I don't know how it
measured up in general instruction mix terms, as I never worked on one.
3. Does KLH10 support any of the XKL extensions?
4. Do the freeware copies of TOPS-20 floating around
the internet have
any support for these extensions or were those proprietary to the
copies of TOPS-20 that shipped with TOADs?
The extensions are XKL proprietary as of this writing.
Thanks for answering this barrage of silly questions,
Not silly at all. I'm glad that you asked.
Server Engineer, PDPplanet Project
505 5th Avenue S, Suite 900
Seattle, WA 98104
mailto:RichA at vulcan.com
(206) 465-2916 cell