[Next] [Up] [Previous] [Home] [Other]

The Case of the Bashful Computer

Several models of the IBM 360 computer have had drawings of their front panels included on these pages so far.

Some models of the IBM System/360 were more common than others. The larger, more expensive models, such as Model 75 and Model 91, were relatively rare. Finding pictures of the front panel of the System/360 Model 75 was difficult, while pictures showing the front panel of the Model 40, for example, in sufficient detail to make out the legends on the individual lamps on the front panel are relatively common.

But there was one computer of the front panel of which I could find no images at all on the Internet: that was the System/360 Model 85.

This computer was a very important computer historically, as it was the first computer sold commercially to include a cache memory. Some previous computers did include fast storage for the last few instructions fetched, to allow high-speed loops, but a full-fledged cache, which included data as well as instructions, and which even allowed data to be written to the cache, had not previously been used in computers.

With the STRETCH computer, and to a lesser extent in the System/360 Model 91, IBM ran into difficulties with superpipelining. Their mainstream computers, such as the IBM 7090, did overlap instruction fetch, decode, and execute, but superpipelining, where the execute phase of an instruction is broken up into several small pieces had failed to yield the expected gains in performance due to considerations such as branch prediction.

In the case of cache memory, however, the gain in performance it provided for the Model 85 exceeded expectations, with cache misses turning out to be rarer than feared.

The success of cache memory led IBM to re-implement the Model 91 design in monolithic integrated circuits, which had also been used for the Model 85, but with a cache memory similar to that of the Model 85 (but larger and with some improvements) to produce the System 360 Model 195. For a very large scale system, it was quite successful, and the design was updated by the addition of new instructions to produce the System 370 Model 195.

Going back to the question of the front panel design of the Model 85, even taking that step which is necessary when one wants real information, going to a library and looking at real books (or, in this case, an old issue of Datamation magazine) only yielded limited information.

But that limited information still disclosed an important fact.

One of the larger System/370 models, the System/370 Model 165, and the Model 168 which was the same computer with the addition of a few features such as Dynamic Address Translation to facilitate time-sharing use, had a front panel with knobs and switches, but no lights.

Instead, a CRT next to the front panel could be used to examine the computer's registers and carry out others of the tasks of the computer operator or the field engineer.

And, as well, two large microfilm displays were next to, and at right angles with, the front panel, one to display certain documentation about the computer, the other to display an overlay over a set of front panel lights.

Thus, it was desired to display so much information about the computer's internal state that a conventional front panel would simply have been too large.

What the advertisement revealed was that the System/360 Model 85 had the same type of microfilm display units. They were cream in color, rather than black, as befitted the different color scheme of the earlier machine.

Later, Al Kossow added the manual from IBM entitled System/360 Model 85 Functional Characteristics to his site, and with the complete layout of knobs and switches on the Model 85 front panel, I was able to prepare the following diagram:

The top half of the diagram illustrates the front panel of the Model 85, the bottom half that of the System/370 Model 165. Some aspects of the illustration of the System/360 in the diagram above are still conjectural.

As you may have noticed, the front panels have quite a bit in common in addition to the presence of microfilm displays and a CRT.

One of the things they have in common is the numeric keypad on the front panel, used to enter addresses and data. This, of course, long predates either the Interdata 8/32 or the Digital Equipment Corporation PDP-8/A.

Ed Thelen's site, another very informative site about early computers, had a section with a number of IBM product announcements for IBM customer engineers that I had only recently noticed. One for the System/360 Model 85 was present there.

I had known that the microcode for the Model 85 was in words that were 108 bits long, as this figure is given in a standard work on computer microcode that gives comprehensive details about the microcode formats of the IBM System/360 models 40 and 50, as well as those of the RCA Spectra/70 model 45 and the Honeywell H4200, Microprogramming: Principles and Practices, by Samir S. Husson, published by Prentice-Hall.

The product announcement stated that the Model 85 computer had read-only storage containing 2,048 108-bit words of microcode, and that it also had writeable control store for diagnostic purposes, expandable from 512 108-bit words to 1,024 126-bit words. Actually, it said 515 108-bit words; I think that's a misprint. But wouldn't the 126 also be a misprint?

When I discovered that the 360/85 had the microfilm panels like those of the 370/165, I had asked about whether or not the similarities ran deeper, and someone noted that the microcode word of the latter was not 108 bits long, but 126 bits long.

Doing some web searching about the 370/165 and 370/168, I turned up the claim that the IBM System/370 Model 3033 computer was somewhat hastily produced, after IBM's ambitious Future System project was abandoned, and was largely based on the IBM System/370 model 168 design. (It did not, however, have a front panel in any way resembling that of the 370/168; in fact, it only had a small vestige of a front panel.)

There were some innovations in the 3033 design, however, that make it significantly more than a repackaging of the 168. For example, its execution unit, unlike those of its predecessors, included a limited degree of pipelining within itself in addition to the overlap between the I-unit and E-unit characteristic of the series. The microcode also was significantly rewritten to improve its performance.

On Al Kossow's site, the manual IBM 3033 Functional Characteristics is also present. Unlike the corresponding manuals of an earlier era, this manual went as far as to disclose that this computer had 3,072 words, each 108 bits long, of read-only control store, and 1,024 words, each 126 bits long, of writeable control store, used for diagnostic purposes.

Thus, it seemed like the recollection one person had of a microcode word length for the 370/168 that differed from that of the Model 85 could now be explained; both of those machines, and the IBM 3033 as well, used microcode words 108 bits in length to implement their normal instruction set, and microcode words 126 bits in length for special diagnostic functions.

This would seem to imply, as well, that all three machines used essentially the same microcode format, making the System/360 Model 85 microarchitecture a particularly long-lived microarchitecture that served IBM well, despite the fact that sales of the System/360 Model 85 itself were disappointing, due to economic difficulties at the time of its introduction.

As it happens, the microcode format of the 3033, at least, is described to an extent in William Stallings' Computer Organization and Architecture.

Also, it might be noted that all of IBM's work on its Future Systems project did not go to waste. Eventually, quite a bit of what had been envisaged for that project wound up in IBM's Application System/400 line of computers.

[Next] [Up] [Previous] [Home] [Other]