Talk:IBM System/360 Model 67

Plan

I am currently revising/creating a series of articles related to the early days of CP/CMS. This involves the various systems referenced by the family tree at the end of this article, plus entries on Cambridge Scientific Center, Lincoln Laboratory, Type-III product, etc. Kindly bear with me while I get the various pages in synch. (I was doing some of this work in a private sandbox, but thought it would be better to start putting out some text. I do intend to provide a good number of citations and references, plus do some systematic editing/wordsmithing, over the next few weeks. Obviously feel free to jump in and fix anything that needs attention. Trevor Hanson 22:04, 19 October 2006 (UTC)

How many were made

I may be confusing the model, but is it possible only four were made, and the machine was announced only as a competitive strategy to break another company (perhaps Burroughs). IBM never plannned on it being a production machine, just a give away for universities. --ArmadilloFromHell 22:12, 19 October 2006 (UTC)

No, sorry, there were very many S/360-67s sold. Every CP/CMS site initially ran a S/360-67. National CSS had perhaps a dozen at one point. (IBM did take a lot of heat about the announcement of the S/360-67 and how long it took before they were shipped; and of coure the TSS/360 operating system was a washout, and this led to a lot of issues. But it was certainly a real system. I think CDC scrapped the most over the -67, but in the end Honeywell won Project MAC – the main sales effort the -67 was targeted to win.)

I'm not sure what other system you may be referring to. There were never many S/360-91, S/360-95, or S/360-195 systems; they were definitely real machines, but essentially research/experimental machines.

Can you provide any more details? We can try to track this down. I have Pugh's 800-page tome IBM's 360 and early 370 systems sitting at my desk here and I'm sure everything is there. Trevor Hanson 22:32, 19 October 2006 (UTC)

I probably have the wrong model number. My vague recollection is that for one model of 360 when announced, the machine did not even exist, the photograph was a fake mockup. It was announced in response to another company creating a new machine, and was intended to stop people buying that machine. The then had to scrample to create a real machine to avoid looking stupid. In the end they made only four and having achieved the marketing ploy, made no more. I think it was a machine designed more for "time-sharing" rather than batch processing. --ArmadilloFromHell 00:17, 20 October 2006 (UTC)

I have subsequently remembered more about this. In hindsight I should have figured it out right away. The "phony" system installed at only a few sites was the S/360-91. It was a real machine, but it was used in a phony way by IBM to compete against the CDC 6600. This led to a bitter legal battle which was eventually settled in 1973. There was a famous incident with a mocked-up front panel used for photos; I can't find it cited anywhere, but I remember hearing about it. The most well-known S/360-91 was probably the one at Columbia, which you can see here (http://www.columbia.edu/acis/history/36091.html) and which was a monster. There is some more about the antitrust situation here (IBM antitrust). IBM apparently lost something like $110M on the model, and there's a good argument that they didn't INTEND to use it in a phony way; their marketing rhetoric just got ahead of their delivery capability. Who knows? Trevor Hanson 00:36, 17 November 2006 (UTC)

Multiprocessor support

What were the features that provided better multiprocessor support? I vaguely remember reading about a feature of the Model 65 (and probably 67 as well) where references to the first N Kbytes of memory (for some value of N that I don't remember) would refer to different regions of physical memory on the different processors; were there any M67-specific features (other than ones needed only for systems with VM) for MP support? Guy Harris 21:54, 28 November 2006 (UTC)

Another good question. I will need to do some research. My update was based on the "blue letter" text, which claimed special MP capabilities for the -67 without going into detail. I have a 360-67 green card somewhere here; if I can find it I'll look for some MP-oriented instructions. I do see the following note in Pugh's book:

The first MP support for S/360 was an OS/360 extension called Attached Support Processor (ASP). Available in March 1967, it supported two processing units linked by a channel-to-channel adapter rather than by shared memory. The first...with shared memory was OS/MVT Model 65 Multiprocessing, available in March 1969.... LCS could not be attached in this configuration.

But I also just found this statement: "Planned for later release [in TSS] was software support for additional languages and for a multiprocessor version of the model 67 with up to four processing units." I'm not sure if this implies that the fourplex wasn't actually built. I seem to remember hearing about at least one, perhaps at FAA or NSA or something. NCSS might even have had one. I'll hunt around. In the meantime I'll reword the statement. Trevor Hanson 04:20, 29 November 2006 (UTC)

There is evidence in System/360 Model 67 Time Sharing System Preliminary Technical Summary (1966) that IBM was considering offering as many as four processors with the model 67. See page 28: “The IBM 2846 Channel Controller is contained in a single stand-alone frame and has its own power supply. It permits interconnection of multiple I/O channels to multiple processors and multiple core storage units. Seven I/O channels, four processors, and eight storage units may be interconnected by one channel controller. Four controllers can be attached to a System/360 Model 67. The I/O channels attached to the channel controller are the 2860 and 2870. Since a total of seven channels can be attached to one controller, TSS can have as many as 28 channels.” Note, however, that the IBM 2365 processor storage was limited to four memory busses, thus limiting the total number of processors plus channel controllers to four. John Sauter (talk) 21:44, 12 August 2010 (UTC)

I agree. The University of Michigan Computing Center had a S/360-67-2 (half duplex), followed by a replacement S/360-67-2 (full duplex). The original one CPU system was a half-duplex so it could be field upgraded to full-duplex, but that didn't work out and a new full-duplex S/360-67-2 configuration had to be installed in 1968. Later the UM Computing Center wanted to add more than two processors and I was told by the Director of the UMCC at the time (R. C. F. Bartels) that IBM came back with a price for the upgrade, but the price was so high that he felt that it was clear that IBM didn't really want to provide the upgrade (they were probably hip deep in the work for the new S/370 models around this time). MTS, the time-sharing operating system developed for the S/360-67 at Michigan, was designed to run on up to four processors, but only ever ran on one and two processor configurations of the Model 67. Jeff Ogden (talk) 22:12, 12 August 2010 (UTC)

I suspect the price included development costs for an eight-port IBM 2365, an expanded version of the IBM 2846 channel controller to handle four processors, and a new model of IBM 2167 which doubled its size. The “quadriplex” configuration wouldn't have been supported by TSS, but I doubt University of Michigan would have cared about that. John Sauter (talk) 11:48, 13 August 2010 (UTC)

The basic features used by OS/360 M65MP were

• A modification to the behavior of the Set System Mask (SSM) instruction
• Direct Control in order to allow each processor to interrupt the other
• Prefixing, which provided a prefix register on each processor to redirect references to page 0

The 360/67 improved on this because it allowed each processor to access all of the I/O channels directly, eliminating the need for shoulder taps. Note that OS/360 did not exploit this, although TSS/360 did. Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:45, 3 September 2010 (UTC)

Thanks - prefixing was the feature I was thinking of. Guy Harris (talk) 20:49, 4 September 2010 (UTC)

DAT vs. the "Blaauw Box"

This article includes the following statement which I'm not sure is correct and which may be misleading:

The S/360-67 design included a radical new component for 
implementing virtual memory, referred to as the "Blaauw Box" 
after its designer Gerry Blaauw. This device was originally 
designed during the main S/360 project, but had been excluded 
from the S/360 design; it was revived for IBM's failed proposal 
to Project MAC, for a customized S/360, and finally came to 
fruition in the S/360-67. 

The following quote is from the article "The IBM 360/67 and CP/CMS" by Tom Van Vleck that is available at http://www.multicians.org/thvv/360-67.html:

CP-67 didn't use the segmentation features of the 360/67; 
In fact, segmentation was not a feature of the original Blaauw 
proposal, but was derived from a 1966 JACM paper by Arden, 
Galler, O'Brien, and Westervelt of Michigan. Both TSS and 
MTS did use segmentation, but not as pervasively as Multics 
did on the GE 645. 

Tom Van Vleck's description agrees with the memories of two staff members (Mike Alexander and Scott Gerstenberger) who worked at the University of Michigan's academic Computing Center when the first 360/67 to be delivered outside of IBM arrived at Michigan in the fall of 1966. Van Vleck's description also agrees with Bruce Arden's unpublished memoirs. Bruce was an Associate Director at UM's Computing Center and was directly involved in the discussions with IBM about the model 65M (what became the model 67). He is also one of the authors of the JACM article that Van Vleck mentions.

I'd suggest rewording the current article to read:

The S/360-67 design included a radical new component for 
implementing virtual memory, the "DAT box" (Data Address 
Translation box). DAT on the 360/67 was based on the 
architecture outlined in a 1966 JACM paper by Arden, Galler, 
Westervelt, and O'Brien and included both segment and page 
tables. The Model 67's virtual memory support was very similar 
to the virtual memory support that eventually became standard 
on the entire S/370 line.

Jeff Ogden (talk) 06:18, 5 July 2010 (UTC)

I went ahead and made the change suggested above. Jeff Ogden (talk) 01:48, 15 July 2010 (UTC)

Rework to focus more on the 360/67 and less on issues covered in other articles

This article is or should be primarily about the IBM S/360 Model 67 computer, but it also talks quite a bit about CP/67 and Virtualization. Unless there is an objection, I plan to rework the article to sharpen its focus on the 360/67 and to add links to other articles on CP/67 and Virtualization. I will also try to address the issue of too few inline citations. Jeff Ogden (talk) 01:56, 15 July 2010 (UTC)

The changes suggested above were made last week. Jeff Ogden (talk) 16:38, 5 August 2010 (UTC)

Configurations

Under “Features” I found this sentence: “A simplex configuration allowed from 2 to 4 processor storage units (512K to 1M bytes). A duplex configuration allowed from 3 to 8 processor storage units (768K to 2M bytes).” While strictly speaking this is correct, it is misleading because there were three basic configurations available: simplex, half-duplex and duplex. The half-duplex configuration could have as few as two IBM 2365 components. If there is no objection I will add this. John Sauter (talk) 21:27, 10 August 2010 (UTC)

No objection from me. I think this is a good change. Jeff Ogden (talk) 01:38, 11 August 2010 (UTC)

I have reorganized the “Features” section into three parts: “Features”, “New Components” and “Basic Configurations”. John Sauter (talk) 13:54, 11 August 2010 (UTC)

Use of Large Core Storage

At least one installation, Carnegie Tech, had 2361 Large Core Storage (LCS) on a 360/67 used with a modified TSS/360. I don't know whether there was any official IBM support for that. Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:50, 3 September 2010 (UTC)

There wasn't. The IBM 2361 was not supported on the System/360 model 67, though I believe it would have worked on the simplex configuration. It would not have worked on the half-duplex or duplex model 67, or on the model 65 in multiprocessor mode, because those systems used multi-ported IBM 2365 memory, and the IBM 2361 did not have a multi-port model. John Sauter (talk) 22:21, 4 September 2010 (UTC)

You're mistaken; not only did it exist but it was written up. Lauer, Hugh (1967). "Bulk core in a 360/67 time-sharing system". 1967 Fall Joint Computer Conference. AFIPS Conference Proceedings. Vol. Volume 31. Academic Press. pp. 601–609. {{cite book}}: |volume= has extra text (help); Cite has empty unknown parameters: |laydate=, |laysummary=, |trans_title=, |month=, |trans_chapter=, |chapterurl=, and |lastauthoramp= (help); Unknown parameter |separator= ignored (help)CS1 maint: extra punctuation (link) CS1 maint: postscript (link)

The article does mention an intent to upgrade to a duplex configuration; I don't know whether that was ever done. Shmuel (Seymour J.) Metz Username:Chatul (talk) 02:57, 5 September 2010 (UTC)

When I wrote “there wasn't” I meant there wasn't official IBM support for the IBM 2361 on a System/360 model 67. See IBM. IBM System/360 System Summary (PDF). GA22-6810-12. {{cite book}}: Unknown parameter |separator= ignored (help)CS1 maint: extra punctuation (link) page 6-20. I am not surprised to learn that the IBM 2361 did work on the simplex System/360 model 67, but making it work on the duplex configuration would have required serious re-engineering of the IBM 2361. John Sauter (talk) 13:27, 5 September 2010 (UTC)

That would depend on what official is. Normally when IBM sells a large system they require a system assurance analysis to ensure that the pieces will work together. If it passes system assurance and IBM agrees to maintain it then I would consider that to be official support even if it is an RPQ. There are certainly other options that IBM sold but didn't include in the System Summary, e.g., the SUMP unit on the 2040. Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:13, 5 September 2010 (UTC)

Is there any evidence that Mr. Lauer's system was under IBM support? It would be simplest to ask him—is there anybody here who knows him? He is listed in [Linkedin]. (By the way, thanks for the reference to the AFIPS paper; I have added it to the IBM 2361 Wikipedia article.) John Sauter (talk) 19:21, 5 September 2010 (UTC)