|
http://ed-thelen.org/comp-hist/philco-212.html
Philco 212 of February 14, 2006 from Ray Gollub [below] stirs excitement in me to remember a world of so long ago where we worked together—especially when he referred to me as the father of the main frame design. I did experience the birth of Philco’s computer entry into this exciting pioneering activity and the joy of working with such dedicated and capable people.
I joined Philco in the spring of 1953 and was assigned to their Government and Industrial Division in Philadelphia, Pa. Shortly after accomplishing some minor assignments I was asked to evaluate the Norden Bomb and Navigation System as used by the US in World War II, and suggest some possible improvements. The Norden system used equipment that had evolved over a period of time from 1940 to 1945. It was last implemented with an electro mechanical computer with major use of ball-disk integrators. Its major limitation was finding sufficient highly skilled machinists to handle the volume of units that might be needed in case of another large scale war and also the reliability and ability to withstand stress with a vacuum tube implementation would prove to be not acceptable.
Soon after I started at Philco a threesome from Bell Laboratories, William Shockley, John Bardeen, and Walter Braittan gave a talk at the Engineers Club in Philadelphia describing their development of the transistor. I was greatly impressed with the possibilities. So was Bob Noyse of the Lansdale Division of Philco who started a program to start producing some. My officemate Ralph Brown began experimenting with them and invented the Direct Coupled circuit which was ideal for implementing logic circuits. I decided on a parallel and asynchronous structure based on the Institute for Advanced Study design. The project went ahead with the proper interface with the airplane and a small magnetic disk to store the programming and temporary results. The equipment was stored in a metal container that was the same size as the mechanical model we had received. The system was delivered to the Navy, installed and flight tested.
Representatives from National Security Agency visited us and expressed interest in our transistor computer, and provided a specification of each of the functions they required. I studied their specifications and met with them to review precisely each of the functions required, some of which were unique to NSA’s needs. They invited me to their facility to meet John Eakus who asked me to program a number of examples on their current computer to further test my understanding. I made a perfect score and they were ready to go ahead with the project. We named it the Transac S-1000 (SOLO), and delivered it along with a Programmer’s manual in November of 1957.
Our next venture was to supply a computer to the Bureau of Ships which we referred to as the Transac S-2000. This is the one that Ray Gollub said had experienced a financial problem. The financial problem was resolved and the computer was delivered and moved to the Navy department that made use of it.
During this same period we bid and won the contract to build a super rugged computer for the US Army to be mounted on the back of a truck that could withstand the rigors of excessive bouncing in the field.
The Transac nomenclature was replaced with numbers such as Philco 210, 211, and 212. By 1960 we were designing the Philco 211 and our small group of cubicles moved from the old Atwater Kent building on Wissahickon avenue in Philadelphia to a large modern Building in Willow Grove and we became the Computer Division of Philco. We provided Philco 211s to a number of users which included Westinghouse, General Electric, US facility under the Cheyenne Mountain that kept account of satellites, and the government of Israel.
While installing a Philco 211 computer at the Bettis division of Westinghouse we realized that the computer required about eight hours to come to a possible iterative solution to a three dimensional partial differential equations. Computer speed was of the essence when running iterative programs. Bettis was developing nuclear power capability for US submarines. I had a session with Dr. Wolf to review the mathematics. It became clear that adding a single combination ‘multiply-add’ instruction would help. We also found a faster add by operating on four bits at a time rather than one. The pre fetching mentioned by Ray Gollub also helped.
The designers of the 212 enjoyed the use of an automated logic design language rather drawings on mountains of paper needed to describe the many interconnections between/among registers. This was beneficial not only to the designers but also any trouble shooting needed. The software for this was designed by Kathe Jacoby.
Ray Gollub’s team finished the design, implementation and testing of the Philco 212 ahead of schedule. The urgency of the customer however was such that they started using it on Philco’s premises.
|
|
When I joined Philco in late 1957, their "computer lab" was working on a
navy contract to provide a transistorized mainframe designated the "CXPQ".
While this machine was delivered, I believe that it was never made
operational, as the funding ran out. However, Philco used this design to go
into the scientific computer business.
The developers included many dedicated and brilliant people, a few of whom
I'd like to mention. The "father" of the mainframe design was James L.
Maddox. The tape transport development was led by George Cogar, who later
founded Mohawk Data Systems, the seminal key-to-tape system. The tape
subsystem controller design was led by "Neil" Eldert. The printer subsystem
design was led by Monroe Postman and Harriet Bien.
Only a few model 210 systems were built, as faster transistors became
available, and, at the suggestion of Professor Morris Rubinoff (Ruby) of
the University of Pennsylvania, the model 211 replaced the 210. The 211 was
the same mainframe with MADT (micro-alloy, diffused-base) transistor
circuits substituted for the SBTs. I think the improvement in speed was
about fifty percent.
As for the claims of being asynchronous: It's true that these mainframes had
no system clock, but, except for the adder, were not truly asynchronous.
The timings for other mainframe operations were determined by the
propagation time from one control state to the next. If that period proved
insufficient, pairs of inverters were inserted into the path.
These mainframes initially found good acceptance in the nuclear reactor
design industry and at least one aircraft manufacturer. The users group
was known as "TUG" (Transac User's Group), which refused to change the
acronym when Philco stopped using "Transac" in favor of "Philco 2000"
systems.
The basic programming tool was ALTAC, Philco's Fortran compiler. The
initial version of ALTAC, of course, had its share of bugs. The revision,
ALTAC II, was well accepted. (Professor Saul Rosen consulted on programming
matters, and Dr. Louis R. Lavine led the elite programming team.)
The competition, of course, was the IBM 7090, which seemed to have been
rushed into existence on an accelerated schedule when the Philco machines
appeared. When the CDC 1604 showed up, that didn't help Philco either.
Early on, the need for faster core storage was evident. A project was
initiated to develop a two microsend memory. Unfortunately, this project
was badly mismanaged and eventually outsourced.
I think it was about 1960 when we started to work on the next model, the
212. While I was nominally in charge of that effort, and contributed some
of the key decisions, the guiding mind clearly was that of a gentleman named
Richard P. Brown, a recent hire. IBM had announced the STRETCH system, in
which the system was using look-ahead techniques to improve performance.
Dick Brown borrowed from that idea and proposed a simplified architecture,
using pre-fetch of instructions and a separate register to hold operands
being stored, permitting the earlier execution of the next instruction. The
specifications on speed, using 8k modules of two microsecond memory, was 4x
that of the 211. This specification was met.
The three models, of course, were upward program-compatible. I recall that
the 212 incorporated some hardware enhancements affecting the programming
which, if reflected in the code generated by ALTAC, would not, of course,
run on the earlier machines. I don't recall that this was ever an issue,
since a simple specification in the compiler would generate appropriate code
for the target system.
The competition for the 212 was the CDC 3600. While the two machines, I
think, were of comparable performance, and most of the pricing was
comparable, CDC's memory was far cheaper, contributing to Philco's marketing
problems.
After Ford bought Philco, and while the prototype 212s were being debugged,
the computer division was put into the hands of Dean Wanlass, formerly, I
believe, of Packard-Bell. I left Philco in 1962 or 1963, after seeing the
initial 212 accepted. By that time, as may have been expected, Wanlass had
filled many of the key positions, including those in engineering, with
relatives and acquaintances.
In a post-departure interview at Philco corporate hq, I was told in plain
language that the computer division did not figure in Philco's long-term
plans. Although work at the computer division continued for another year or
so, it was indeed closed out.
I would welcome any comments or corrections to this little piece of computer
history.
|
