Talk:IBM System/360/Archive 2

Archive 1

Archive 2

No Computer Pictures

None, not one, of the pictures on the current article shows the computer. They show the peripheral devices. The gallery of the Model 50 is perfect: it shows the console (today the screen), the printer, and the keyboard, but not the the computer. Just as a desk PC today is housed in a featureless box so was the 360.
See the photo at this link The 360 computer is in the featureless red boxes in the bottom half of that picture. ALL of those boxes are necessary to complete the computer. In the photo it is in 50 featureless red boxes each bigger than a standard refrigerator. FIFTY!! Really.
I first saw a working 360 in 1968, and last saw one in 1995. I am sure of this. I think the problem is that today it is hard to believe how big the 360 was.
Does someone have a public domain photo of the computer? I do not.
Best to all, Nick Beeson (talk) 01:57, 17 February 2016 (UTC)

What do you mean by "the computer"? The Big Red Box in the middle of the System/360 Model 30 picture, with the control panel on the front, is the IBM 2030 Processing Unit, which is the machine that contains the system registers, ALU, microcode, and core memory. See the IBM 2030 Processing Unit, System/360 Model 30, Field Engineering Manual of Instruction. Perhaps that picture doesn't show all the peripherals, but it most definitely shows the entire CPU (no, it doesn't have the side panels off to show you all the circuitry, but the CPU circuitry is all in that box).

In the smaller models, the data channels were implemented with the same hardware and microcode that implemented the CPU. In the larger models, the data channels were separate boxes, so you might have multiple boxes for "the computer" (i.e., everything but the peripherals), and might have some or all of the main memory in separate boxes. Guy Harris (talk) 02:27, 17 February 2016 (UTC)

So what if there are no pictures of "the computer" whatever that means. The article is about the IBM System/360, a system which consists of a lot more than just "the computer". I'm tempted to remove all the "computer not shown" additions as irrelevant since there are lots of things not shown but I'll wait and see what other editors think. Tom94022 (talk) 06:59, 17 February 2016 (UTC)l

I'd certainly remove them. The first coupleI looked at were pictures of the front panel and CPU, which would certainly seem to be "the computer". Saying "computer not shown" is misleading at best. Peter Flass (talk) 14:49, 17 February 2016 (UTC)

I've already fixed the captions on the Model 20 and Model 30 pictures, as the FE manuals for those models clearly indicate that the box with the front panel is also the box with the CPU in it. I don't have any manual to prove it, but I strongly suspect that the IBM 2065 Processing Unit is the box with the front panel, so that the CPU is what's pictured in the Model 65 image as well, although the data channels are in separate boxes. Guy Harris (talk) 18:35, 17 February 2016 (UTC)

They are now all gone. The captions are correct; none profess to be an image of a "computer" but most in fact have at least a part of the box containing the processing unit. The only exception is the 91 console which I am pretty sure did not contain the processing unit, but the image doesn't assert it is a image of the computer so the negation is superfluous and now gone. Tom94022 (talk) 19:41, 17 February 2016 (UTC)

Channel Architecture

There is no mention of the channel architecture. It is sufficiently complicated that it probably use its own article. I am probably qualified to write such an article for the pre-1985 channel. I don't know if the channel architecture should be considered something of lasting importance. I know that it has changed with fibre optic cables etc. In the 3081 the relatively simple view from the main-frame (a small number of instructions (Start I/O, Halt Device, etc) was replaced by complex data structures which shared between the OS and the CPU.

IMHO the idea of retaining the DMA storage address within the CPU/channel rather than storing it in the device as in UNIVAC 1107 ESI and many DMA schemes seems more secure. The idea of a channel command code which was partially decoded in the channel: write, read and read-backwards and fully decoded in the device was an improvement over the 7090 channel scheme. Rdmoore6 02:17, 4 January 2006 (UTC)

I think an article on IBM's channel architecture would be very valuable. --agr 19:32, 7 February 2006 (UTC)

Be careful not to mention Seymour Cray or the Peripheral processors on his CDC 6000 family of the previous year, or people will see where IBM copied this idea from. T-bonham (talk) 17:01, 6 April 2010 (UTC)

That's doubly wrong. First, the PP's on the CDC 6600 aren't remotely similar to the I/O channels on the S/360. Second, the channel architecture on the S/360 is clearly inspired by that of the IBM 7000 series, which preceded the CDC 6600. Those channels, in turn, were patterned after the Exchange on the IBM 7030 Stretch, which was designed in the 1950's. Shmuel (Seymour J.) Metz Username:Chatul (talk) 22:32, 29 August 2010 (UTC)

I'm not aware of any security issues with the UNIVAC ESI scheme, which was also used on the UNIVAC 490. I see the most notable feature of the S/360 channel architecture as the standardization of the CCW command codes.

I would suggest separate article for

• The Exchange on the 7030
• The 7607 and similar channels on the IBM 700/7000 series
• The 7907, 7908 and 7909 channels on the IBM 700/7000 series
• S/360 channels Shmuel (Seymour J.) Metz Username:Chatul (talk) 22:32, 29 August 2010 (UTC)

Would this be the right place to mention interleaved storage? I don't think it has appeared anywhere in the article. Alternatively, it could be mentioned in the discussion of main storage? Zvmphile (talk) 05:06, 19 November 2015 (UTC)

IMO this is neither the appropriate article nor talk section for interleaved storage. I'm not even sure it is notable but someone did see: Interleaved memory. Tom94022 (talk) 07:16, 19 November 2015 (UTC)

A section discussing I/O channels would definitely not be the right place to mention interleaved storage. As Tom94022 noted, there is an article about the general concept of interleaved memory; it might be worth mentioning briefly that some models had it, perhaps indicating which ones did (for example, in the table of models), with a link to interleaved memory. Guy Harris (talk) 07:41, 19 November 2015 (UTC)

Information on interleaving belongs in IBM System/360#Table of System/360 models; it does not belong in a section on I/O. Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:59, 19 November 2015 (UTC)

The interleave factor should be in the various tables. I don't know that it was special for S/360, but the faster machines using slow core use it to get the speed needed. I am pretty sure that CDC and Cray also used it. Many machines now will do two way interleave if you put in matching SIMMs, maybe four. The 16 way interleave for the 360/91 is somewhat rare. Gah4 (talk) 06:24, 12 May 2016 (UTC)

PCP

"PCP was used on intermediate machines" - I guess if you look at the whole line, wasn't it /40 and above? This was the low-end of machines that could run OS. Was PCP ever much used except for running special-purpose stuff such as ASP support processor code? Peter Flass (talk) 00:56, 12 May 2016 (UTC)

As far as I know, PCP would run on a 64K 360/40, using 20K and leaving 44K. PL/I (F) was designed to be able to run in 44K. If needed, it will keep the symbol table on disk. For smaller machines, you would use DOS/360, and for larger ones, you might run MFT, and run MVT for even larger machines. PCP might also have been used for sysgen, though later OS/360 discontinued PCP and used MFT for sysgen. Gah4 (talk) 06:12, 12 May 2016 (UTC)

I only knew ASP running on larger machines, such as the 370/168 and 360/91. There is also a stand-alone RJE station that IPLs directly from cards. No OS needed. I believe it was usually run on a 360/20, but could run on other models. (Conditional assembly to select 20 or not.) Gah4 (talk) 06:12, 12 May 2016 (UTC)

Attached Support Processor ran on two machines - the main machine and the support machine. Peter Flass was talking about ASP running on top of PCP on the support processor, which was a smaller machine such as a 360/40, not a /50 or /65 or /75 or /91 - those were the main machines. Guy Harris (talk) 06:41, 12 May 2016 (UTC)

Removed spurious 2's complement assertion

I removed the following incorrect contention, in section "Influential features":

The System/360 introduced a number of industry standards to the marketplace, such as: [...] Two's complement arithmetic

Its quite inaccurate to claim the System/360 introduced (or even standardized) 2's complement in computer architecture. The use of 2's complement representation had been used in numerous computers which predated the System/360, such as the EDSAC and is discussed in the First Draft. 2's complement has numerous advantages in binary computers, which is why nearly every binary computer uses it.

I'm not sure EDSAC is part of the computer "marketplace", but the DEC PDP-6 preceeded the IBM System/360, used two's-complement binary (even for floating point!) and was definitely a commercial machine. I am embarrassed that I did not notice the error myself, since I programmed the Stanford PDP-6. John Sauter (talk) —Preceding undated comment added 04:26, 6 December 2009 (UTC).

As did the earlier PDP-4 and PDP-5. --Brouhaha (talk) 11:35, 6 December 2009 (UTC)

How about popularized twos complement? Many of the IBM predecessors, such as the 7090, are sign magnitude. Univac and CDC were selling ones complement machines at about the time of S/360. Just as with the 8 bit byte, others may have done it earlier, but S/360 got people thinking about it more. Gah4 (talk) 02:24, 17 February 2015 (UTC)

I'm dubious about the PDP-4, which according to our article is a cheaper PDP-1. The PDP-5 was not a commercial machine and only 23 PDP-6s were sold, again according to our articles. IBM dominated the market and the S/360 made two's complement a de facto industry standard.--agr (talk) 08:19, 17 February 2015 (UTC)

The PDP-4 supported both 1's complement and 2's complement addition, as per the programmer's manual (the "add" instruction did a 1's-complement addition, the "tad" instruction did a 2's-complement addition). The "opr" instruction could complement the AC, but not complement it and add 1 to the result, so no single instruction would negate the AC. It probably wasn't the machine that popularized 2's complement, even if it could somewhat support it.

In any case, S/360 certainly was not the first line of computers to use 2's complement; it may, however, have been the first mainstream computer to do so, and thus perhaps the machine that made it the industry standard. Guy Harris (talk) 08:51, 17 February 2015 (UTC)

I am bothered by the implication that only IBM produced “mainstream” computers. There were other computer manufacturers before April 7, 1964. Just because IBM dominated the market doesn't mean that only their computers deserve to be called “mainstream”. I think the use of two's complement binary representation for integers was already the industry standard by 1964, and IBM was catching up. The use of 8 bits for characters, on the other hand, was not. IBM decided it would be best to use 8 bits for a character so they could maximize the number of 4-bit digits that would fit in a given amount of memory. That was new, and the industry followed. John Sauter (talk) 12:28, 17 February 2015 (UTC)

I'm not going to argue, but only give you my recollections. Certainly many lines of computers were "mainstream" in the 60s, and it wasn't clear that one manufacturer would eventually dominate. However, I think that the other mainstream computers used either 6-bi BCD with 36-bit words (GE, UNIVAC, IBM 70xx), 48-bit words (Burroughs) or 18-bit words. Later most machines seem to have kept the 36-bit word using 9-bit bytes. The other flavor of machine, like the 14xx, H-200, etc used 6-bit BCD but was not word-oriented. I think, although I would be willing to be shown to be wrong, that the 360 was the first to combine the two concepts. Was it the first to use a 32-bit word with 8-bit bytes? Probably nothing to do with twos complement, except that sign-magnitude was also popular in many systems. Peter Flass (talk) 13:00, 17 February 2015 (UTC)

I'm bothered by the inference that only IBM produced "mainstream" computers, as that certainly wasn't what I was trying to imply. I'm not sure the PDP-{4,5,6} were "mainstream" back then. If 2's complement was the industry standard by 1964, what other machines were using it? GE was, UNIVAC wasn't Burroughs wasn't (sign-magnitude floating point in the B5000, as I read the manual), RCA was using it in the process-control RCA 110 computer (manual) but not in the RCA 301 (training manual) or 501 (manual), NCR doesn't appear to be using it in the 315 (manual), and CDC was, as far as I know, using 1's complement. Guy Harris (talk) 18:48, 17 February 2015 (UTC)

It takes time for new ideas, even after they have become accepted in the industry, to be implemented in products. There is always a tendency to maintain compatibility with the old products, in order to hold onto your existing customers. I regard the PDP-6 as a mainstream computer, even though not very many were sold—some very respectable institutions purchased them, and used them for real work. It was accepted at the time that newly-defined instructions sets would use two's-complement, since it was clearly better than sign-magnitude or one's-complement. Nobody was surprised that IBM's new instruction set architecture used two's-complement. I didn't know GE was using it, though that doesn't surprise me. The Burroughs B5000 was designed in the late 1950s and 1960, so it may have predated the realization that two's complement was better than sign-magnitude. John Sauter (talk) 22:55, 17 February 2015 (UTC)

Were the institutions in question universities or research organizations? If so, they could be respectable and still not buy mainstream computers, as they're almost expected to buy weird machines. :-)

So perhaps 2's complement was the standard in computer company engineering departments at that time, but not the standard in the marketplace, in which case IBM wasn't catching up to, say, RCA or NCR or UNIVAC in the "mainstream" computer market (their 2's complement machines appear to have come out after S/360); it's more like they were "middle of the pack" (behind GE for mainstream computers, but ahead of or, at least, parallel with, the rest of the Seven Dwarves). Guy Harris (talk) 23:27, 17 February 2015 (UTC)

The only PDP-6 computers I have touched were the Stanford, MIT and Rutgers machines, and they were both universities and research organizations. Stanford and MIT were certainly expected to buy (or build) wierd machines, but I wouldn't put Rutgers in that category. In addition, according to the list of PDP-6 customers at PDP-6 serial numbers there were also PDP-6s which were used for commercial timesharing. John Sauter (talk) 15:23, 18 February 2015 (UTC)

The PDP-6 is pretty much the same family as the PDP-10, and later models of the PDP-10 were reasonably mainstream. Otherwise, the 704, 709, 7090, 7094 series were kept sign magnitude to keep back compatibility. The 360 allowed IBM to ditch back compatibility, and switch to twos complement. (Except when running emulation microcode.) But why did CDC use ones complement for so long? Gah4 (talk) 06:45, 12 May 2016 (UTC)

IBM 2361

peripheral or component?

I added a reference to the IBM 2361 on June 18, and all references to the IBM 2361 were removed by user Chatul within a day. Is there a concern that the IBM 2361 did not in fact exist? I have invited user Chatul to respond here. John Sauter (talk) 20:20, 20 June 2010 (UTC)

I did not remove the 2361; I moved the reference where it belongs; look for LCS. The 2361 was not a peripheral device as the term is used in the computer industry; it is an adjunct to the processor.

is there a reason that you did not ask for the reason I changed it before reverting the change?

Now, if you believe that the model number should be included in the text, I have no quarrel with that. Shmuel (Seymour J.) Metz Username:Chatul (talk) 21:17, 20 June 2010 (UTC)

I do not agree that the IBM 2361 was not a peripheral device—It was an option which could be attached to System/360 models 50, 65 and 75, and could even be attached to two systems, providing a communications path between them. Granted, it did not attach by way of a data channel, and was not programmed as an I/O device, but it did live in its own cabinet.

I did not ask you for the reason before adding it back to the list of 23xx-class devices because I noticed its omission and thought that I had forgotten to add it. After the add I did some more checking and found that you had deleted it, so I started a topic here to avoid an edit war. Really, you should have started the topic rather than delete material which had been added only hours before. John Sauter (talk) 21:45, 20 June 2010 (UTC)

Actually, I didn't delete it; I relocated it and plugged in the name.

Would you also list the 2860 and 2870 as peripheral units? They were both in external cabinets.

The term peripheral equipment has a long established meaning in computer hardware, and it doesn't refer to external cabinets. A disk drive in the same enclosure would still be classified as peripheral equipment. Memory has never been classified as peripheral equipment, although it was often in external boxes. —Preceding unsigned comment added by Chatul (talk • contribs) 23:15, 20 June 2010 (UTC)

I agree that “peripheral equipment” has a long-established meaning. Consulting Wikipedia I find the following language: “Peripheral equipment...is any equipment used to give a computer system more features.... Any device for the computer that the computer can work without is peripheral equipment.” I believe this is the correct definition, and yes, that does mean that the IBM 2860 and 2870 are peripherals, since the System/360 Model 65 can operate without them. John Sauter (talk) 23:49, 20 June 2010 (UTC)

The IBM Mainframe. April 7, 1964: IBM transforms computing IBM introduces the System/360™, a family of five increasingly powerful computers that run the same operating system and can use the same 44 peripheral devices.

Every IBM publication that I have checked uses the term peripheral consistently with that cite.

I note that you didn't list the 2365 as a peripheral device. How does its status differ from that of the 2361? Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:28, 21 June 2010 (UTC)

While there were many peripheral devices that could be used by all models of System/360, there were also some that could not. For example, the IBM 1404 was supported only on System/360 models 30, 40 and 50. Do you have a definition of the term which differs from the one I cited above? The IBM 2365 is on the boundary between being a peripheral and being a necessary part of the system, like the CPU. The first memory box is required, all subsequent ones are optional, and hence peripherals. John Sauter (talk) 12:34, 24 June 2010 (UTC)

IBM has consistently used the term peripheral device to refer to I/O equipment. I haven't found a single S/360 manual that uses the phrase in any other sense. Shmuel (Seymour J.) Metz Username:Chatul (talk) 10:02, 25 June 2010 (UTC)

But the IBM 2361 is I/O equipment! It can be used to transfer data from one System/360 to another, just like the channel-to-channel-adapter (CTCA). Surely you would agree that the CTCA is an input-output device, like any other communication controller, such as the IBM 2701 or 3705. Just because the IBM 2361 is on the memory bus rather than the channel bus shouldn't disqualify it as an I/O device. Consider the early DEC PDP-11s, which used a single bus for both memory and all other I/O devices. John Sauter (talk) 21:42, 25 June 2010 (UTC)

I know of no vendor that considered RAM memory to be I/O equipment, and certainly IBM never has.

As for the CTCA, it is physically attached to I/O channels at both ends and is driven with channel programs.

Were the index registers on the 650 I/O devices? They too were in a separate box. Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:23, 27 June 2010 (UTC)

The IBM 2361 is physically attached to the memory bus of both systems (when it is shared) and is driven by CPU instructions. Does the choice of bus, or the programming method, define an I/O device? With reference to the IBM 650, if the index registers were necessary to the functioning of the system, I would not call them peripherals. This is similar to the power supply or the console of the IBM 7090, which were separate units but required. John Sauter (talk) 15:54, 27 June 2010 (UTC)

The Memory bus is not an I/O channel. The IBM 2361 has the same status as the IBM 2365; it's just slower.

The index registers on the IBM 650 are necessary to the functioning of the system when an instruction specifies indexing; the IBM 2361 is necessary to the functioning of the system when an instruction is present in LCS. Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:23, 27 June 2010 (UTC)

The memory bus can be used as an I/O channel when the IBM 2361 is connected to two systems. The ability to share an IBM 2361 between two systems is an important difference from the IBM 2365. Yes, the IBM 2361 is necessary when it contains an instruction to be executed, but likewise a card reader is necessary when it contains an instruction to be executed (as when doing an IPL from the card reader). That doesn't mean that the card reader is not an I/O device, or not a peripheral. A part of a computer is “necessary” when any meaningful use of the computer requires the part. Thus, the CPU and some memory are necessary, but a card reader, CTCA, or LCS are not.

Perhas by your definition of I/O channel, but I know of no hardware vendor that referred to a processor-addressable memory as an I/O channel, not even on machines with memory-mapped I/O. Shmuel (Seymour J.) Metz Username:Chatul (talk) 09:14, 28 June 2010 (UTC)

Is it possible to make meaningful use of an IBM 650 without using the index registers? Will the machine function with the index registers removed? If so, I would have to regard the index registers as an optional part of the system, and therefore as peripherals. John Sauter (talk) 18:57, 27 June 2010 (UTC)

So on a machine that uses external memory boxes and does not require memory at location 0, you would call all of the memory peripheral equipment because the machine could function with any one of the memory boxes removed? The processors in an MP are peripheral equipment because the machine could function with any one of them removed? Shmuel (Seymour J.) Metz Username:Chatul (talk) 09:14, 28 June 2010 (UTC)

full name

In the IBM System/360 model 65 functional characteristics manual, the IBM 2361 is called both “IBM 2361 Core Storage” and “IBM 2361 Large Capacity Storage (LCS)”. Examples of both can be seen in the table of contents. It is never called “IBM 2361 Large Core Storage”. If there is no objection, I will correct the name in the article. John Sauter (talk) 12:47, 30 June 2010 (UTC)

See T. A. Humphrey, "LCS (Large Core Storage) Utilization in Theory and in Practice", Proceedings of the Spring Joint Computer Conference, 1967, AFIPS, p. 719 {{citation}}: Unknown parameter |separator= ignored (help) Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:38, 30 June 2010 (UTC)

T. A. Humphry appears to have been a researcher at IBM Houston in 1967 and 1968. In his paper he references the IBM SRL: GA22-6869-1 2361 Core Storage Original Equipment Manufacturer's Information. I contend that Mr. Humphry knew the component as the LCS (which is the overwhelming majority of his references to it in his paper) and guessed that this acronym stood for Large Core Storage. I believe he was mistaken; based on the System/360 model 65 manual, LCS stood for Large Capacity Storage. I have been unable to locate the manual for the IBM 2361 on-line, but I believe that the model 65 manual is a more authoritative source than Mr. Humphry's paper, even though he was an IBM employee. John Sauter (talk) 04:25, 1 July 2010 (UTC)

I'd consider a researcher who worked at IBM to be more relevant than a nontechnical manual. Now, if you have a CE manual for the 2050 or 2065 that refers to the 2365 as Large capacity storage, I'd consider that to be relevant. I did find Computers in Spaceflight: The NASA Experience, - Chapter Eight -, - Computers in mission control -, Manned mission control computers

I tried to find a copy of IBM System/360 Operating System Introduction to Main Storage Hierarchy Support for IBM 2361 Models 1 and 2. C27-6942, {{cite book}}: Unknown parameter |separator= ignored (help) with no luck. I'll have to check whether I have it on a dead tree. Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:53, 1 July 2010 (UTC)

I would not regard any IBM SRL as “non-technical”. All the references I have been able to find to “Large Core Storage” are from NASA Houston and IBM Houston. However, since that name was apparently used over several years for the component, I suggest we call it by all three names: “IBM 2361 Core Storage”, “IBM 2361 Large Capacity Storage (LCS)” and “IBM 2361 Large Core Storage (LCS)”. The first two names are in IBM SRLs, the third was used by NASA and IBM Houston. John Sauter (talk) 21:43, 1 July 2010 (UTC)

Having heard no objections, I have added the three names for the IBM 2361. I have also corrected the name of the IBM 2365, and added stub pages for both, giving additional information about these components. John Sauter (talk) 19:33, 3 July 2010 (UTC)

Shared memory is pretty popular in current systems, and never that I know considered an I/O device or peripheral. Many systems use memory-mapped I/O, such as PDP-11 and VAX, but I/O devices are still considered peripherals. Seems to me that the main distinction of the 2361 is that it is slower than non-LCS memory. But not as slow as I/O devices. Gah4 (talk) 06:55, 12 May 2016 (UTC)

Strong compatibility statement

I found this statement was recently added to the article:

“Full backwards compatibility for System/360 software is available until present day with the IBM zSeries computers.”

Taken literally, I seriously doubt that this is true. The z/Series computers are organized so differently from the original System/360 computers, that I doubt you could IPL an old magnetic tape of OS/360 and run applications under it. Much more believable would be the claim that compiled OS/360 applications could be run without recompilation under z/OS. In any case the statement needs a reference to support it. Does anyone know what compatibility is offered by the z/Series? —Preceding unsigned comment added by John Sauter (talk • contribs) 14:49, 26 November 2010 (UTC)

The statement says software compatibility so I do not quite see where you see the problem. It might be difficult to find the old tape reader but afaik all software would work unmodified. Anyway the same claim is in the introduction of the IBM System z article as I read it. A little googling reveals this:

• http://arstechnica.com/hardware/news/2007/07/legacy-matters-why-the-ibm-mainframe-continues-to-thrive.ars
• http://www.processor.com/editorial/article.asp?article=articles%2Fp3130%2F32p30%2F32p30.asp

Btw I would think sure could you run an ancient copy of OS/360 unmodified, not on the bare hardware but in a partition. The hypervisor and partition logic itself is builtin into System/360 since the beginning - it could always run several different OS in one computer virtualised. Richiez (talk) 17:33, 26 November 2010 (UTC)

Thank you for the references. The first reference supports the claim that z/OS provides application program compatibility back to System/360, but makes no claim that the old operating systems would run on a new CPU. The second reference makes a vague claim of full compatibility, but without details or citation. I know that the ASCII bit was removed from the PSW with the introduction of System/370, and I vaguely recall that the I/O architecture was re-done at some point, perhaps even removing the original four I/O instructions.

You were correct initially. There are incorrect assumptions in Richiez's comments.

First, you can not run OS/360, or any S/360 operating system, on current IBM hardware. As part of System/370 Extended Architecture (S/370-XA) IBM replaced the S/360 and S/370 I/O instructions with instruction for a new channel subsystem that did some of the housekeeping formerly done by the operating system; this continued with Enterprise Systems Architecture (ESA) and z/Architecture. Initially IBM provided customers with the ability to set a processor or LPAR to S/370 mode, but no current processor has that capability.

Second, the hypervisor and partition logic itself is not builtin into System/360; the first IBM processor with PR/SM was the 3090. You could install CP-67 on the 360/67 and VM on the S/370, but they did not provide the ability to run S/370 mode operating systems on a box that did not support S/370 mode. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:18, 26 November 2010 (UTC)

You pretty much can't run OS/360 on any 370 with the 4K key feature(sic). If it ever sets two different keys to the same 4K block, something will fail soon after. Gah4 (talk) 07:11, 12 May 2016 (UTC)

As far as running an old operating system not on the bare hardware, that isn't compatibility but emulation. I can run OS/360 on an emulator on a PC.

But not directly in an LPAR, or directly in z/VM.

I think the original statement should be modified to read as follows: “Full backwards compatibility for System/360 application software is available with the IBM zSeries computers.” John Sauter (talk) 19:03, 26 November 2010 (UTC)

Even that is too strong. Maybe

Backwards compatibility for System/360 application software is available with the IBM zSeries computers, with some caveats:

• There is no ASCII mode. (ASCII mode was not included in System/370 since IBM was unable to identify any System/360 user that had found a use for the capability provided. Dpbaird29 (talk) 18:27, 1 December 2010 (UTC)) (I am not sure this is correct; is there a citation available? John Sauter (talk) 20:22, 1 December 2010 (UTC))

None of the standard operating systems supported the ASCII bit. I would have expected IBM to simply assume that no customer used it, rather than to conduct a formal survey. For purposes of the article the lack of support is probably enough, but I'd be interested in any documentation of the stronger claim. Shmuel (Seymour J.) Metz Username:Chatul (talk) 21:50, 1 December 2010 (UTC)

• There are no specification interrupts for halfword and fullword instructions.
• Some old OS facilities are gone. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:18, 26 November 2010 (UTC)
• Some OS behaviors have changed

References

Thanks for the clarifications. So the binary pattern for a few instructions has been recycled and hence those won't even raise an exception?

I'm not aware of any opcode recycling between S/360 and S/370, although there certainly was opcode recycling later, e.g., the opcodes for vector instructions. What I was referring to was specification interrupts. On all S/360 models but 85 and 195, the instruction

    LH    R1,7

would cause an interrupt because 7 is not halfword aligned. On a S/370 it would load the bytes at 7 and 8 into R1 and extend the sign. Shmuel (Seymour J.) Metz Username:Chatul (talk) 23:56, 29 November 2010 (UTC)

Regarding the hypervisor mode for System/360 it was my oversight when I mistook it for 370. Incidentally I believe there must have been hypervisor functionality available for 360 (most likely only internally at IBM) during the development of System/370 because I remember articles claiming that the hypervisor and system software was tested before the availability of the actual 370 hardware. Richiez (talk) 11:36, 27 November 2010 (UTC)

It's not true for S/370 either. PR/SM came in with the 3090. There was hypervisor functionality available earlier, but it was via operating systems that you had to install, not one that came preinstalled into the licensed internal code of the machine. Shmuel (Seymour J.) Metz Username:Chatul (talk) 23:56, 29 November 2010 (UTC)

Errors in Note #1

This note contains a commonly held misunderstanding of the relationship between the development of System/360 and external standards. I had the privilege of joining the System/360 development team at a fairly early stage (see Frederick P. Brooks, Jr, The Design of Design (2010), p.328 for a reference to my participation in the design competition). A formal system of communicating with serially numbered confidential memos was used to communicate within the efforts in the Poughkeepsie, Endicott and Hursley, England laboratories. Distribution of these memos was tightly controlled to ensure that Department of Justice concerns that information about unannounced products was not used in a noncompetitive way and that competitors did not learn of IBM's plans. As far as I know, copies of these memos were retained until the records retention order of the judge in the DOJ antitrust action was raised and the records destroyed. Based in the information contained in Pugh, Johnson and Palmer's IBM's 360 and Early 370 Systems, I suspect that these memos are no longer available. I authored some of the memos and received a number of the memos written and am writing a revised version of the note based on my memories of the development activities. I am including citations to information currently available on the internet. Others memories of what happened in the early '60s and I look forward to seeing their comments and corrections. --Dpbaird29 (talk) 22:19, 24 November 2010 (UTC)

Note 1 relates to ASCII versus EBCDIC, which wasn′t obvious from your comment. Could you explain what the misunderstanding is?

It would be helpful if you could propose alternate text or, COI permitting, edit the article, or at least the one note. Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:33, 25 November 2010 (UTC)

I'm new to editing Wikipedia and added this entry before editing the article. I find that I've a bit to learn on including citations, etc. I'll get back to this after Thanksgiving. Please excuse this newbe. Dpbaird29 (talk) 14:31, 25 November 2010 (UTC)

The note in question is unsourced, so we'd be happy to get some guidance from someone who was on the inside. I took a look at note 1 and I found several problems from what I know. The passive construction "System/360 was originally to use the ASCII character set" is problematical on its face for a large organization like IBM. Who made this initial requirement? IBM management, the system architect, some engineer? Nor did they decide "instead" on EBCDIC. My early copy (A22-6821-1, June 1964) of IBM System/360 Principals of Operation says (p. 11-12) that the system "can handle any eight-bit character set ... However all character-set sensitive I/O equipment will assume either" EBCDIC or ASCII-8. Also there was no one BCD system, as note 1 implies. Finally, "EBCDIC had been used in some earlier systems" seems dubious to me. All earlier IBM systems used 7-track tape drives (6 bits per character) and except perhaps for the 7030, none had 8-bit internal character facilities. --agr (talk) 16:32, 25 November 2010 (UTC)

The decisions were made well before IBM published Principles of Operation. What does the SPREAD report say about ASCII and EBCDIC? Also, while the I/O architecture was neutral, the decimal arithment feature was designed around ASCII-8 and EBCDIC.

The SPREAD report was written in December 1961.[1] The ASCII standard came out in mid 1963. Principles of Operation documents the design of the first System/360 products announced 1964. That design supported both ASCII and EBCDIC, not just EBCDIC. The note implies otherwise.--agr (talk) 19:32, 26 November 2010 (UTC)

All earlier systems except Stretch used only 7-track tape drives. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:14, 26 November 2010 (UTC)

SPREAD provided the strategic objective of developing a completely compatible 5-processor product line. A major debate in the design competition that Fred Brooks discusses on pp. 75&76 and 313-328 in the book cited above, was the tactical choice between a 6-bit byte or an 8-bit byte. "The future application promise of the lowercase alphabet was convincing to me. We settled on 8-bit bytes, ..." [Brooks (2010), p. 319] Both IBM's 6-bit Binary Coded Decimal (BCD) and its 8-bit EBCDIC character code standards included multiple code pages to accommodate roman alphabets of more than 26 characters. There was do internationalization of ASCII when System/360 was announced. An effort was made to reconcile the ASCII printable characters with those of USA EBCDIC. This was generally successful, however, ASCII ended up with a (broken) "vertical bar" (that I find on my keyboard) and EBCDIC ended up with a "logical OR" represented by a solid vertical bar. Unfortunately Pugh, et al did not include this topic in their "definitive reference work on the early history of System/360..." I'd appreciate comments on whether (and where) some of this history should be made available on Wikipedia. I was attempting to keep Note 1 as brief as possible. Dpbaird29 (talk) 18:08, 29 November 2010 (UTC)

I think a sourced section on the history of the 8-bit character decision belongs in this article. It's one of the longest lasting legacies of the 360. --agr (talk) 19:15, 29 November 2010 (UTC)

It may be difficult to source all the historical items that I consider significant. The System/360 Architecture chapter in Brooks (2010)helps but doesn't get into the ASCII-EBCDIC debate. Pugh, et al (1991)were not involved in the early stages of System/360 design and didn't research this area for their book. Dpbaird29 (talk) 14:20, 2 December 2010 (UTC)

What are the the historical items that you consider significant and which are attributable to the sources you have? Maybe we can help.--agr (talk) 21:10, 2 December 2010 (UTC)

The ASCII-8 code in the S/360 Principles of Operation isn't the ASCII that everyone else uses, and ANSI never standardized it. Otherwise, last year I set the A bit in the PSW on a 360/20, executed some decimal instructions, and got the appropriate sign codes. It does work! I do wonder how much microcode it takes to convert between punch codes and EBCDIC bytes. Gah4 (talk) 07:18, 12 May 2016 (UTC)

Horizontal and Vertical Microcode?

On February 20 there was an edit which added references to “horizontal” and “vertical” microcode. I have no idea what this means; some explanation, or a reference to an article providing an explanation, is needed. Unless someone objects, or an explanation is provided, I will revert the edit. John Sauter (talk) 05:42, 22 February 2012 (UTC)

There's a reasonable (if brief) explanation of these terms at Microcode#Implementation. The material in the lead doesn't make much sense, though. "While the more powerful machines added a vertical (software implementation) microcode layer" seems confused about the distinction between software and vertical microcode; the two are similar, though not the same. Software implementation would be closer to PALcode on the Alpha, or to 80x87 (or other FP) emulation packages on other architectures. With some exceptions, they would ordinarily be less powerful than implementations closer to the hardware. In the absence of a source to clarify these points, I think deleting this material might be the best approach. Jakew (talk) 10:08, 22 February 2012 (UTC)

Unfortunately I can't find a chart in Pugh that summarizes this, or much of a discussion. He does list the control store type by S/360 model, but that isn't relevant to this discussion. /30: Card Capacitor (CCROS), /40: Transformer (TROS), /50 and /65: Balanced Capacitor (BCROS), /75 and /95: None. He doesn't discuss horiz vs. vert. Peter Flass (talk) 13:12, 22 February 2012 (UTC)

Looking at John Savard's model descriptions (http://www.quadibloc.com/comp/pan04.htm) it's obvious from the word sizes that all S/360 models had horizontal microcode only. Peter Flass (talk) 13:17, 22 February 2012 (UTC)

I didn't think the edit was correct since it is my recollection that all S/360 processors used mostly horizontal microcode but I have a vague recollection that some instructions in some machines were interpreted with "vertical" microcode. I haven't had time to research it but based upon Peter Flass representation above and the lack of reference I would not object to reverting the addition. Tom94022 (talk) 16:53, 22 February 2012 (UTC)

FWIW "All but two of the initial 360 models (the high-end Models 75 and 91) were microprogrammed" and the word lengths make them "horizontal". If there are any IBM'ers reading this one should take a look at S.G. Tucker, "Microprogram Control for System/360," IBM Systems Journal, v. 6, n. 4, 1967, pp. 222-241. (The JRD's are no longer free :-)) I'm going to revert the edit Tom94022 (talk) 07:13, 23 February 2012 (UTC)

The word length an RAM is not relevant to whether the processor is horizontally or vertically microprogrammed. What is relevant is the length and structure of the microinstruction. The 360/75, 360/91, 360/95 and 360/195 were hardwired. The 360/40, 360/50, 360/65, 360/67 and S60/85 had horizontal microcode. I don't have access to the CE manuals for the 360/22, 360/25 and 360/30, but my recollection is that they were vertically microcoded. Would it be TMI to include some of the CE manuals in the references? Shmuel (Seymour J.) Metz Username:Chatul (talk) 22:51, 1 March 2012 (UTC)

If you found a place in the body to add each CE Manual as a reference supporting a statement they would not be TMI but that may be a bigger task than you want to take on. I'm not an expert on WP, but it might be OK to just list them in See Also under a sub section Customer Engineer Manuals. Tom94022 (talk) 05:33, 2 March 2012 (UTC)

The horizontal vs. vertical distinction is common in discussions of microcode, though that might not have been until after S/360. Definitely the 360/20 is vertical. I suspect I would have called 360/40 microcode vertical, but it isn't so easy to say. The boundary isn't sharply defined. I do remember being surprised when I learned that the 370/168 used microcode, before I understood the idea of horizontal microcode. It is common, but incorrect, to associate microcode with slow machines. Gah4 (talk) 07:28, 12 May 2016 (UTC)

Leasing vs buying

The second paragraph notes that before the S/360 people were reluctant to buy computers. To recall history, prior to about 1964, you couldn't buy an IBM computer since the company only leased them. They lost a court decision before changing that policy. I started to edit the article to describe this, but may simply alter the paragraph rather than complicate the article with this discussion. Any comments? Lou I 18:55, 23 Jan 2004 (UTC)

Second most expensive project

System/360 is described as the second most expensive project of the 1960s (after Apollo). SAGE is estimated to upto 12 billion 1964 dollars by some. Drhex 13:59, 2004 Aug 26 (UTC)

UWA 360

No longer at Shenton Park - in storage at the Australian Computer Museum Society WA warehouse

Search for UWA University Computer Club

Extremely Vague Imprecise Speed Claims

"The slowest System/360 models announced in 1964 ranged in speed from 0.0018 to 0.034 MIPS;[2] the fastest models were approximately 50 times as fast[3] " -- as fast as what? Are the fastest models 0.0018 * 50 MIPS to 0.034 * 50 MIPS = 0.09 MIPS to 1.7 MIPS? How much more vague and inaccurate could you possibly get? Has history actually recorded how fast the fastest System/360 actually was, or it the actual number actualy unknown and disputed, with historians still arguing about the speed, some claiming it was as low as 0.09 MIPS, and others claiming it was as high as 1.7 MIPS? Was the actual number, classified, so IBM could only report an extremely wide range of possible speeds and not get into specifics? What gives? Why not simply use the actual numeric values for the range of top speeds, instead of requiring the reader to perform interval arithmetic to calculate such an vast range of possible values? Xardox (talk) 14:35, 3 October 2014 (UTC)

I've tried to clarify the text. Speeds then as now can be squishy, there were no standard benchmarks then, and this is further complicated by the fact that the fastest IBM 360s listed in the original 1964 announcement were never delivered, real models came a few years later.--agr (talk) 16:34, 3 October 2014 (UTC)

Discussion of speed can be sticky; in many cases you get widely different results depending on how you measure things. In the particular case of the S/360, IBM has published some fairly complicated timing formulae.

Meaningless speed claims is usual, especially for high-end machines. For the 360/91, with out-of-order execution and instruction overlap, times depended very much on the actual instruction mix and execution order. The goal was to execute floating point one operations per clock cycle, though it rarely did that, with the 750ns core a bit reason why. In actual use, the 360/91 was reasonably comparable to the 370/168 for floating point, a little slower for non-floating point. Close enough that you could submit jobs without knowing which one it would run on. Gah4 (talk) 07:16, 15 May 2016 (UTC)

The article mentions the 360/91, which replaced the originally announced 360/92. There was also a 360/95, which used the same processor but a faster memory.

The later 360/85 was faster than the 360/95 on some workloads but slower than the 360/91 on others. The 360/195, however, was uniquivocably faster than the 360/91 and 360/95.

Note: http://bitsavers.org/pdf/ibm/360/funcChar has manuals for the 360/85 and 360/195, but I don't see the 360/95. I did, however find the IBM 360/95 press release . Shmuel (Seymour J.) Metz Username:Chatul (talk) 19:40, 5 October 2014 (UTC)

As well as I know, the 95 is the same processor as the 91, but with faster thin-film memory. That might be close enough not to get its own Functional Characteristics. Gah4 (talk) 07:16, 15 May 2016 (UTC)

This is not so much my area of expertise so I'll defer to the judgement of Seymour and Arnold but here are some thoughts:

• Pugh says the initial 360 (M30-M75) achieved a performance range of about 25:1 and then goes on to describe separately the high end achievements and problems in Chapter 7, in particular see Table 7.1. Perhaps you can take the same approach, citing Pugh in two brief steps, announcement (M30-M75, ignoring 60, 62, etc) achieving a 25:1 performance range and then supercomputing (85,91, 195), 3 to 7 times the 75 depending upon mix.
• Is there perhaps too much detail in the lede about the M91?

Just my 2 cents :-) Tom94022 (talk) 00:20, 7 October 2014 (UTC)

No, the initial S/360 was M30-M70; the M75 came later and replaced the earlier M70 (and M72?). Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:41, 24 May 2016 (UTC)

I've tried to clarify that the large memory capacity discussion is not just about the Model 91. I'm reluctant to cite a speed ratio because it is so dependent on which machines and when. The 91 is a good representative of the top end close to the announcement. The 195 came later and was more a preview of the System 370, with true "monolithic" integrated circuits, instead of the hybrid modules used in the rest of the 360 line. The model 65 was perhaps the most popular high end machine, but it's not listed on the IBM history site.--agr (talk) 03:22, 7 October 2014 (UTC)

The original System/360 announcement included processors with a performance range of 25 to 1. Six years later this had increased to around 200 to 1, and today the performance of the 3081 is approximately 450 times that of the System/360 Model 25.

A. Padegs, "System/360 and Beyond," IBM JRD, V. 25 N. 5, Sep 1981

Rather than putting specific numbers in the lede how about paraphrasing this reliable source and leaving the details to the table? Tom94022 (talk) 18:12, 11 October 2014 (UTC)

Wasn't it the /50 that was about 1 mips? Maybe this could be used to convert relative speeds to abslute. Peter Flass (talk) 19:33, 11 October 2014 (UTC)

Computer Speeds From Instruction Mixes - pre-1960 to 1971 This suggests the M50 was slower, but we are talking about content of the second paragraph of the lede, not what's in the body of the article. There were at least 12 models of the S/360 without counting sub-models so why have any one or two numbers in the lede when the important point for the lede is that one breakthru feature of S/360 was a family of compatible processors with a 25-to-1 range of performance at announcement which was increased to 200-to-1 over the next four years. This is in contrast to the current, "...the Model 30, could perform up to 34,50 ..." now in the lede. The performance of a specific model can be in the section of the article relating to a model or in the table we might add or both. Tom94022 (talk) 22:10, 11 October 2014 (UTC)

Model 92

Recently, a line was added to the model list for the 92. No announcement date is given, and no ship date. The comments state that it was replaced by the model 91. When the models 64 and 66 were added I challenged them, asking for a source. A source was found, and those models are well-attested. Is there a source for the model 92? If not, I will remove it. John Sauter (talk) 05:10, 28 October 2016 (UTC)

The model 92 now has a reference, but as I read it (pages 390–391) the core memory version of the model 92 was replaced by the model 91, and the thin film memory version was replaced by the model 95. Also, page 386 says that a model 90 was named in a footnote of the original System/360 announcement. Was the model 92 ever named outside of IBM? If not, perhaps we shouldn't list it. On the other hand, since the model 90 was described publicly, if only in a footnote, perhaps we should include it. If there are no objections, I will change 92 to 90, and say that it was replaced by the models 91 and 95. John Sauter (talk) 11:27, 28 October 2016 (UTC)

I found a reference to the model 92 in the March 3, 1980, issue of ComputerWorld, page 72. It says that the model 92 was announced in August of 1964, and IBM gives the date as August 17. Also, the ComputerWorld issue of October 8, 1979, page 17, mentions that the model 90 was replaced by the model 92. If there is no objection I will add the model 90 as announced with the other early models, replaced by the model 92 announced August 17, 1964, and replaced in its turn by the models 91 and 95. John Sauter (talk) 12:17, 28 October 2016 (UTC)

I added the reference for the Pugh book. Page 387-390 have a pretty detailed description. The change from 90 to 92 happened when addition improved to two cycles, and multiply to three. It was then mentioned at the Fall Joint Computer Conference in October 1964, and compared against the CDC 6600. The 92 was designed for 0.5 microsecond core. To better compete against the 6600, in price and speed, some wanted a cheaper, slower, version, and so the 91, with 1 microsecond core was named, and with the same 75ns cycle time as the 91. In April 1965, the core memory in design for the 60, 62, and 70 improved from 1000ns to 750ns, leading to the switch to the 65 and 75. Not so much later, the cycle time for the 91 and 92 went down to 60ns. At about that time, the 120ns thin film memory went to the 95, and all core memory versions were called 91. As far as I know, there was no 500ns core. The actual 91 runs with 16 way interleaved 750ns core so, with some luck, one access per machine cycle. Oh, also, the 91 and 95 were the first use if monolithic (integrated circuit) memory in a production computer: the storage protection keys are stored in an array of 16 bit bipolar SRAM that can match the 60ns machine cycle. Gah4 (talk) 16:41, 28 October 2016 (UTC)

I have added the model 90. John Sauter (talk) 04:37, 30 October 2016 (UTC)

System/360 models

Section IBM System/360#Table of System/360 models. Some information are not correct (for example Model 20 April 1966 instead of March 1966). See System/360 Dates and characteristics--Tim Stamper (talk) 22:20, 16 October 2017 (UTC)

So you're saying Pugh et al got it wrong, or that somebody misread it? (My copy's stashed away somewhere, and I'd have to dig a bit to find it. Dear MIT: I'd pay for an e-book version, so I can stash it away on my MacBook Pro and always have it available....) Guy Harris (talk) 22:38, 16 October 2017 (UTC)

Some dates are different (see Official IBM website). Model 91 is probably wrong. Have a nice day!--Tim Stamper (talk) 22:58, 16 October 2017 (UTC)

I keep my Pugh handy - the table agrees with Pugh. The Model 91 Wikipedia Article incorrectly cited Pugh - p. 394 states, "The first shipped System/360 Model 91... in late October 1987... ". I trust Pugh over a current IBM archive website; personally I have found a number of errors on those sites. BTW I was having a good day until I had to waste time on this :-( Tom94022 (talk) 00:11, 17 October 2017 (UTC)

Nobody forced you to edit the Model 91 article.--Tim Stamper (talk) 00:19, 17 October 2017 (UTC)

That's right, no one forced me - when I find a discrepancy and I have the time I reseach and resolve it - you apparently just prefer to throw stuff on a talk page and see what happens. Tom94022 (talk) 07:40, 17 October 2017 (UTC)

I saw that the dates are different here, for example: Model 91 announced 1966 (IBM Official website) instead of 1964 (Wikipedia). If I edited the article, you would say: Use the talk page! No more discussion here.--Tim Stamper (talk) 10:58, 17 October 2017 (UTC)

I suspect the problem is with IBM. Announcing something exciting and new gets a big show, and much documentation. Announcing a change in model number of a previously announced model isn't so exciting. It isn't really new, just a new number on something that already (almost) existed. Some might have considered the 91, as a renumber of the previously announced 92, and so not reset the date. Note that they didn't restart the design from the beginning after the number change. Gah4 (talk) 00:25, 19 October 2017 (UTC)

Thanks. Pugh et. al., Emerson (1991). IBM's 360 and Early 370 Systems. MIT. p. 394.: By chance this engineering problem came on the heels of a January 1966 announcement of the Model 91 as a regular product (previously it was limited to special bid, special contract situations).--Tim Stamper (talk) 01:59, 19 October 2017 (UTC)

Actually both announcement dates are valid, November 1964 was an announcement to the sales force; in addition to the above quote we also have from Pugh:

• "… the sales organization asked DSD for a spartan version of the 92 … in November [1964 DSD] agreed to … a system … called the Model 91 …” [p.388]
• "Nomenclature was rationalized in July [1965], when all core-memory versions of the 91 and 92 were renamed Model 91, …” [p.390]

Based upon the above three quotes and Pugh Appendix A I am going to reverse the order in the table, showing Jan 1966 as the announcement and commenting that available on special bid beginning Nov 64. This is consistent AFAIK with all the other announement dates in the table. Tom94022 (talk) 06:21, 19 October 2017 (UTC)

There are four separate citations of Pugh with 7 in line references, so I am also going to consolidate them into one ref with the page number template in line - which means I probably won't get to this until late today or tomorrow, but I will do it. Tom94022 (talk) 16:39, 20 October 2017 (UTC)

Thank you for your edits.--Tim Stamper (talk) 00:49, 21 October 2017 (UTC)

Yr welcome, turned out not to be a waste of time after all. My apologies. Tom94022 (talk) 06:59, 21 October 2017 (UTC)

External links modified

Hello fellow Wikipedians,

I have just modified one external link on IBM System/360. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:

• Added archive https://web.archive.org/web/20110720114849/http://www-07.ibm.com/systems/tw/z/download/clipper_mainframe_at40.pdf to http://www-07.ibm.com/systems/tw/z/download/clipper_mainframe_at40.pdf

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

Y An editor has reviewed this edit and fixed any errors that were found.

• If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
• If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers.—InternetArchiveBot (Report bug) 01:11, 10 November 2017 (UTC)

Found another instance. Guy Harris (talk) 06:47, 10 November 2017 (UTC)

"most complex System/360 model type built"?

IBM System/360#Remaining machines refers to the IBM 9020 as the most complex System/360 model type built. While a 9020 complex was more complex than any single processor, the modified S/360 models making up the 9020 complex were far simpler than, e.g., the 360/91, 360/195. Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:34, 13 December 2017 (UTC)

Agree and changed. There is nothing in any of the references or links to support "most complex." Tom94022 (talk) 19:37, 13 December 2017 (UTC)

Transistor computer

Should the 360 be in List of transistorized computers? Bubba73 ^{You talkin' to me?} 00:02, 16 March 2018 (UTC)

I'd say "yes" given that IBM Solid Logic Technology, which S/360 used, had individual transistor chips. I've added it. Guy Harris (talk) 00:48, 16 March 2018 (UTC)

That was my opinion too, but I don't know much about hardware. The first computer I used was a 360/65 though. Bubba73 ^{You talkin' to me?} 01:47, 16 March 2018 (UTC)

The initial models certainly used SLT, but what about the later 85 and the 195? Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:15, 16 March 2018 (UTC)

The 360/91 (and I presume 195) uses IC memory (16 bit SRAM) for storage keys, as I understand it, the first use of semiconductor ICs in the memory system for a commercial processor. The logic of the 91 (and I believe 195) is ASLT, which is electrically similar to today's ECL, but physically like SLT. As well as I know, the logic of the 85 is still the usual SLT, but the 85 includes cache, unlike other S/360 models. The result was that the 85 could keep up with the 91 on many tasks. Gah4 (talk) 20:41, 16 March 2018 (UTC)

"The Model 195 uses monolithic integrated circuits for the arithmetic and logic operations in the central processor, and as the storage medium in the 32,768-byte buffer memory."

So the 195 1) includes a cache ("buffer memory") and 2) is IC-based. Guy Harris (talk) 21:14, 16 March 2018 (UTC)

Good points about the higher end 360's that should be accounted for in any listing but there is a symantical dispute with @Wtshymanski: at Talk:List_of_transistorized_computers#IBM_360? as to whether the discreet transistors of SLT make it a "hybrid IC" and therefore a computer built with them is not a "transistorized computer." We seem to think at least some models are - perhaps we should take the discussion there? Tom94022 (talk) 17:51, 17 March 2018 (UTC)

References

Article doesn't mention programming languages

This article is incomplete, it does not mention the programming languages used with System/360: mainly Cobol, Assembler and Fortran. Nor does it explain how it was created a batch program for a computer like this: hand written on a sheet of paper, then typed on a punching machine, then the 80-characters paper cards were read into the computer. The only screen display was for the Operator, who directly controlled the machine. It was a bulky computer, requiring a special climatized room with plenty of space for the computer, its operator, the printers, tape readers, dasd devices (disks), punch card readers, etc. The power consumption was considerable. Still a great evolution from the original Turing machines, like Colossus or Eniac. — Preceding unsigned comment added by Martago54 (talk • contribs) 21:31, 15 April 2018 (UTC)

System/360, the hardware, was a separate project from OS/360, the software. (And DOS/360, a separate software projects.) Programming languages should be mentioned in those articles, but I didn't look. Gah4 (talk) 02:37, 16 April 2018 (UTC)

• COBOL (sic), Assembler and Fortran were not the only languages on S/360, and, IAC, belong in the various OS articles, not here.
• There is nothing in the S/360 or its software requiring the use of coding forms.
• The S/360 had 2250 and 2250 screen display devices not part of a console.
• The various S/360 operating systems used keyboard/typewriter and keyboard/display devices to control the system, not just the hardware console.
• Colossus and Enicac were not Turing machines.

Shmuel (Seymour J.) Metz Username:Chatul (talk) 16:51, 16 April 2018 (UTC)

History section suggestion

Please can the history section be expanded to explain the business risks IBM took when developing the System/360. There is a good description here: History of CP/CMS#IBM and the System/360. John a s (talk) 22:40, 22 April 2018 (UTC)

B class?

I haven't checked the criteria, but this looks like a B-class article to me. Bubba73 ^{You talkin' to me?} 07:43, 15 January 2019 (UTC)

I took a quick pass thru the criteria and agree that this is at least a B-class article. Might even be an A-class. Who gets to make the change?Tom94022 (talk) 19:24, 16 January 2019 (UTC)

Any editor can change it up to a B class, so I did. Higher than B (A, GA, FA) takes some sort of review and consensus. Bubba73 ^{You talkin' to me?} 20:20, 16 January 2019 (UTC)

LCM 360/20

As far as I know, the Living Computer Museum's 360/20 is not on display. Last I knew, the 360/30 was in its place, but not (yet) running. Gah4 (talk) 22:58, 21 May 2019 (UTC)

A S/360 M20 does not appear in any search of the LCM site (documents do but no hardware), there are not any S/360 listed in their on-line system status and there are not any S/360 listed in their Mainfram exhibit.

Rather than list a few survivors would we better off to just link the section IBM_System/360#Remaining_machines to World Inventory of remaining IBM System/360 & 370 CPU's.? Tom94022 (talk) 06:22, 22 May 2019 (UTC)

recycling

Most S/360 machines were leased, so went back to IBM. IBM had some tendency to make new things out of them. The 2319 disk drive system is made from recycled 2314s. Old 370/158s were used as channel directors for 303x machines, with channel only microcode. Outside the US, some were sold instead of leased, and that is mostly there they come from today. But otherwise, once IBM got leased machines back, they might have recycled the copper. Gah4 (talk) 06:14, 23 May 2019 (UTC)

Some comments:

• Actually most s360s were short term rented rather than leased. A System/360 could be returned to IBM without penalty on 30 day notice. IBM's lease programs, FTP and ETP, are much more of the S/370 era and beyond.
• Many 360's were purchased particularly by the leasing companies leading to one facet of the 1969 US v IBM antitrust suit. An owner then had dispose of them and in particular the leasing companies had very active re-lease programs. Some end-using owners sold their obsolescent 360s to leasing companies who then released them. But in the end when there was no demand for old iron even at greatly discounted prices the systems wound up salvaged for gold and copper and sometimes spare parts
• IBM rarely made "new things" out of old hardware; the 2319 was a notorious anti-competitive reaction and AFAIK the only disk drive so made "new." In the 2319 case IBM had no new product to match the then superior PCM 2314's and couldn't stand the financial hit of a reduced the rental rate on its large installed base of 2314s so it came out with a "new" 2319 at a much lower lease rate calculated at a price to drive the PCMs out of business while locking in customers until the next generation disk drive came available.
• Like a leasing company IBM tried to re-rent any off rent equipment and then when there was no longer any demand they salvaged it. The one thing IBM rarely if ever did was offer deep discounts on re-renting (e.g. 2314) prefering instead to rent (or lease) newer, next generation, equipment (i.e., the "new" 2319). The leasing companies by way of contrast offered very deep discounts for off lease equipment versus IBM rental rates and then lease rates for the same older equipment. The result is that the purchased 360s stayed in use much longer than the ones IBM rented but in the end they all were salvaged and then scrapped but for a few.

I looked for but couldn't find a RS on the dynamics of the decline of the 360 installations but couldn't find any. I suspect there were three waves or salvaging, IBM off rent, leased/owned technogically obsolete and finally uneconomic due to operational costs. Who knows, there may still be a 360 someplace, probably a government installation, running an old applciation that should be replaced but the government can't or won't appropriate the funds. Tom94022 (talk) 18:06, 23 May 2019 (UTC)