Talk:IBM System/360/Archive 1

Archive 1

Archive 2

Benchmark info?

No info on clock speeds or MIPS in article? How does it compare to eg Pentium?

I don't see any CPU clock rates in System/360 Instruction Timing Information, but the memory speeds vary from 2 microseconds per memory reference on System/360 Model 30, with each memory reference fetching or storing 1 byte (so a 4-byte word would be 8 microseconds) to 1 microsecond per memory reference on the "Model 70" (which never came out, but it's probably close to the Model 75), with each memory reference fetching or storing 4 bytes. A register-to-register add takes from 29 microseconds on the Model 30 to .4 microseconds on the "Model 70".

According to IBM System/360 Model 30 Functional Characteristics, the memory cycle was 1.5 microseconds, with early models having a 2 microsecond memory cycle. It lists the CPU cycle time as .75 microseconds for the 1.5 microsecond memory cycle systems and 1 microsecond for the 2 microsecond memory cycle systems, so that'd be 1.33 MHz for the 1.5 microsecond memory cycle machines and 1 MHz for the 2 microsecond memory cycle machines. The CPU data path in the Model 30 was 1 byte long, so an instruction adding 2 4-byte words would process each of those bytes in a separate cycle. Note also that the general registers were stored in a special chunk of core memory, and access to one of them took 6 microseconds on the 1.5 microsecond memory cycle systems (4 1.5 microsecond memory cycles).

The Model 65 (not the top of the line - that was the Model 75) had, according to IBM System/360 Model 65 Functional Characteristics, a 200 nanosecond CPU cycle time (5 MHz), and a .75 microsecond cycle time for main memory (with a 64-bit memory bus, letting two 32-bit words be fetched at the same time) and an 8 microsecond cycle time for the (again 64-bit) add-on storage unit.

Unlike most modern microprocessors, the main System/360 processors were not pipelined, so it couldn't finish one instruction per clock cycle - instruction N+1 wasn't started until instruction N finished, and instruction N might well take more than one clock cycle. The Model 91 was not only pipelined, but, to a limited degree, superscalar; see IBM System/360 Model 91 Functional Characteristics. It cost many many cubic dollars.

If you were to run the Hercules emulator on a modern PC, it would be a lot faster than even the fastest System/360. I don't know how well it'd do on a Pentium (where "Pentium" means "P5", i.e. Intel's first superscalar x86 chip). Guy Harris 09:46, 5 July 2006 (UTC)

Peripherals

I just added a "Peripherals" section, which includes a one-sentence description of I/O channels ahnd their importance. I consider this section to be a stub, but it is a crucial part of the S/360 story, so much so that a stub is better than nothing. Please improve it if you can.

Do recall that this article is about the S/360 and not about the S/370 or later architectures, at least as I understand it.

Thanks. -Arch dude 00:41, 21 October 2006 (UTC)

Featured article nomination?

Yes, how'bout it? Should we Just Do It? (no reference to expensive high profile brand sneakers here). Or, would the article perhaps need peer review to polish it a little (or a lot)? Tell me what you think, folks. --Wernher 17:25, 17 Dec 2004 (UTC)

The S/360 was the subject of a cover article in Fortune magazine at the time of its introduction, the subject of which was that IBM had bet the company on its new product line. The article included - maybe even in the title - the cost of the S/360. After 44 years I've forgotten what it was, but for the time it was a huge numberAbba1943 (talk) 00:18, 27 July 2008 (UTC)abba1943, 20:19, 26 July 2008

Nice image of an 360

This image of the System/360 Model 91 was taken by NASA sometime in the late 60s.

The following image: http://en.wikipedia.org/wiki/File:360-91-panel.jpg has the description: "This image of the System/360 Model 91 was taken by NASA sometime in the late 60s." ... might be an idea to include it on this page as well? (I personally very much like the image.) I noticed it linked in Front panel. Hfodf (talk) 15:01, 8 February 2009 (UTC)

Done. Guy Harris (talk) 01:25, 23 January 2010 (UTC)

Level of detail for emulator features

I expanded the text to reflect that the emulator support on S/360 was not just hardware. Where is the line between too terse and TMI? I could certainly provide several paragraphs if that is appropriate. Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:37, 21 June 2010 (UTC)

What constitutes a large system?

The 256 KiB cited for large systems was not enough to run either OS/360 MVT or TSS/360 in production. Wasn't 256 KiB actually the minimum memory size for anything large than a 360/50? Shmuel (Seymour J.) Metz Username:Chatul (talk) 12:32, 5 July 2010 (UTC)

Mimimum memory size on the IBM System/360 model 65 was 128 KiB (131072 bytes, using a single IBM 2365 model 1). John Sauter (talk) 13:36, 5 July 2010 (UTC)

New section on architecture

I'm starting a new section on the S/360 architecture. Right now I'm concentrating on getting the information organized, but at there's text elsewhere in the article that at some point should be deleted or integrated into the architecture article. There are a couple of places where the order of presentation is problematical and I will need to either refer to subsequent text or duplicate it in an inappropriate location. I'm quite familiar with the material and am comfortable with the technical aspects, but would appreciate any editorial assistance anybody has time to provide.

Is it reasonable to put in a skeletal section and add text, or would it be better to do the section in user space and only copy it in when it is complete? I've already got text on instruction formats and PSW format, which some might find of interest. Shmuel (Seymour J.) Metz Username:Chatul (talk) 14:17, 6 September 2010 (UTC)

Rather than starting with a new section, start with a new article, and reference it from this one. As the new article progresses, remove duplicated information from this article. When the new article is complete, if the two articles together are not of excessive length, merge them. John Sauter (talk) 16:37, 6 September 2010 (UTC)

Okay, I've started User:Chatul/IBM System/360 architecture. Currently I'm showing PSW bits in a wikitable; I'd appreciate any comments on whether that's a readable approach. Thanks. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:32, 7 September 2010 (UTC)

Note that there's an "Architectural overview" subsection of the "Technical description" section; if this doesn't replace that subsection, that subsection should probably shrink so as not to give too much detail that duplicates your new section. IBM System/370 has an "Architecture details" section; if that duplicates any of what's in your new section, the S/370 section should probably also shrink and refer back to the S/360 section. Guy Harris (talk) 21:50, 7 September 2010 (UTC)

Models 64 and 66

The article contains two references to models 64 and 66, but I have been unable to find any authoritative source for them. Does anyone know of such a source? If there is none, I will delete the references. John Sauter (talk) 05:37, 13 October 2010 (UTC)

DIGITAL COMPUTER hpefthnuitr-,"The IBM System/360 Model 64 and Model 66 four computer processors, eight memories", might be what you're looking for. I remember there being an IBM page that listed all of the models, including those never shipped, but can't recall the URL.

FWIW I do remember the period of "musical models", and it definitely included the 64 and 66, but I haven't retained any of the relevant documents. Shmuel (Seymour J.) Metz Username:Chatul (talk) 09:50, 13 October 2010 (UTC)

The IBM page that lists the models of System/360 is here: System/360 Dates and Characteristics. The newsletter you referenced is dated July 1965, so the undated IBM announcement might have been issued before the April 22, 1965, substitution of the model 65 for models 60 and 62. Note that the newsletter's editorial policy requires submission of material three months in advance of the publication date. Interestingly, the announcement of the models 64 and 66 promised up to four processors and support of the IBM 2361. The model 67 announcement, on August 16, 1965, kept the four processors but eliminated the IBM 2361.

I suspect that the model 64 and 66 announcements were never “official” since they do not appear on IBM's official list of System/360 models, which does include the models 60, 62, 70 and 92. Nevertheless, they should be included in this article, since they are part of the history of System/360. If there is no objection I will add user Chatul's reference to authenticate these models. John Sauter (talk) 14:06, 13 October 2010 (UTC)

Jeff Ogden added another citation to Talk:IBM System/360 Model 67 that mentions a model designation I'd never heard of, 66M. FWIW, I saw the 64 and 66 in various Computerworld articles, not in IBM announcement letters. Shmuel (Seymour J.) Metz Username:Chatul (talk) 09:34, 14 October 2010 (UTC)

Memory sizes for large systems.

IBM's model chart shows sizes of up to 6 MiB without the use of a 2361, and 512 KiB is too small to get good performance with MVT, so I question whether 512 KiB really was a typical size for large systems. Shmuel (Seymour J.) Metz Username:Chatul (talk) 09:46, 14 October 2010 (UTC)

Early model replacements

In the section describing the replacement of, e.g., the 60 and 62 by the 65, there should be mention of the replacement of the 90 and 92 by the 91 and the 95. Shmuel (Seymour J.) Metz Username:Chatul (talk) 21:53, 26 October 2010 (UTC)

I have added a brief note about the model 92. As far as I know there was never a model 90 announced, and the model 95 could be considered the successor to the model 91. —Preceding unsigned comment added by John Sauter (talk • contribs) 04:21, 27 October 2010 (UTC)

more on motivation

There is a little information about IBM's motivation in creating the System/360 set of computers in “a family of computers”, but it begs for more background. Some work has been done on that background in the discussion page of the architecture material, which is still under construction at User:Chatul/IBM System/360 architecture. The full details are interesting, but too long for this article, or even for the System/360 Architecture article, assuming it doesn't get merged with this one. Questions for the group: is this material suitable for an encyclopedia? If so, how should it be presented? Does it deserve its own article? John Sauter (talk) 21:06, 29 October 2010 (UTC)

Related article needs attention

IBM System/360 architecture needs attention and I wanted to invite those watching this page to help. --Pnm (talk) 18:04, 16 December 2010 (UTC)

Merge IBM System/360 architecture into IBM System/360

I've written IBM System/360 architecture with the intent that it eventually be merged here. As part of that, I propose moving or deleting the text in IBM System/360 architecture#Architectural overview and moving some material from IBM System/360 architecture into IBM System/360#Channels. Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:38, 8 November 2010 (UTC)

I don't think merging this into the main IBM System/360 article is a good idea. The main article is already too long and too detailed. I think that the main article is already at risk of "losing the forest for the trees". If anything, the main article needs to be broken up into several sub-articles with brief summaries in the main article that reference the detailed sub-articles. Jeff Ogden (talk) 14:43, 27 November 2010 (UTC)

There's a great explanation of summary style, and I think this is a good time to apply it. It keeps this article from getting too long for a more casual reader, and allows readers to drill down to topics that interest them. I added a {{main}} reference to the architecture section – that section should summarize the architecture article. --Pnm (talk) 18:17, 16 December 2010 (UTC)

Currently IBM System/360 architecture is 58 kilobytes long and IBM System/360 is 47 kilobytes long. Do those sizes mean that a merger would violate Wikipedia rules on article sizes? Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:10, 19 December 2010 (UTC)

There's an editing guideline on article size. It doesn't strictly limit article size, but suggests first considering a split when an article reaches 40 KB. --Pnm (talk) 18:17, 19 December 2010 (UTC)

I removed the "Merge from" template from the article. Jeff Ogden (talk) 12:47, 14 June 2011 (UTC)

Range of Delivery Dates

The delivery date range in the first paragraph of the article was updated recently to 1966 to 1978. However, IBM lists the first delivery date of the model 40 as April of 1965, in http://www-03.ibm.com/ibm/history/exhibits/mainframe/mainframe_FS360.html. I was unable to locate a date for the last delivery of a System/360, but 1978 seems reasonable since the last was “withdrawn” in 1977. If there is no objection I will update the lower bound on the delivery date to 1965. John Sauter (talk) 15:47, 28 March 2012 (UTC)

Bache,et al, "IBM's 360 and early 370 Computer's" lists the 30, 40 and 50 shipping in 1965. If you have a reference that the last “withdrawn from marketing” was in 1978 it is likely the last shipment was that year, since announcing withdrawal can provoke last orders. So I would go with 1965-1978. However, note that such last machines were likely refurbished and not new production. If someone has an IBM Sales Manual or "Service for Consultants" manual from that era it might be possible to figure out the exact year/month. I have one in my garage but don't have the time to go find it. Tom94022 (talk) 23:11, 28 March 2012 (UTC)

There is some evidence that most if not all of the S/360 was withdrawn from marketing before the end of 1977. Google "withdrawn from marketing" with various IBM qualifiers. So maybe 1977 is good Tom94022 (talk) 23:20, 28 March 2012 (UTC)

I have updated the lower bound on the delivery dates to 1965, and added a reference to IBM's web site. IBM agrees with the ComputerWorld ad that the System/360 models 30, 40, 50 and 65 had been withdrawn from marketing by the end of 1977, but there may have been some deliveries after that date. I have left the upper bound of the delivery dates at 1978 pending further research. John Sauter (talk) 12:52, 29 March 2012 (UTC)

Pugh gives 1965: /30-June, /40-April, /50-August, /65-November, /75-January 1966.

Peter Flass (talk) 15:01, 29 March 2012 (UTC)

Remaining Machines

There are a large number hardened versions of the IBM 360 (model 65 IIRC) still in operation on E-3 Sentry aircraft. 31 on USAF E-3's, 1 used exclusively for code updates, 1 used for training and unit status checking and one in Seattle Washinton. 68 total 707 based E-3's were built, but atleast 3 have been lost. The Japanesse 767 version is also running the IBM 360. The remaining U.S. models will be phased out in the comming years with an up comming modification. I post this as I work with the system on the aircraft. Due to issues finding solid sources for this topic I have not updated the article.

• "Airforce-Technology: E3AWACS". Airforce-Technology. {{cite web}}: Cite has empty unknown parameter: |1= (help) —Preceding unsigned comment added by Kesterj (talk • contribs) time, 15:35, 28 May 2010r (UTC)

The IBM System/360 model 65 was a room-sized computer; it would be difficult even to get one aboard a Boeing 707. Could you describe the System/360 you work with? Even a cursory description of the number of boxes and their size would likely pin down the model. John Sauter (talk) 01:09, 29 May 2010 (UTC)

There were airborn System/360 processors not based on any of the commercial models; I don't know whether any of them are still in use or how strictly they adhered to the architecture. Tha name 4 Pi comes to mind. Shmuel (Seymour J.) Metz Username:Chatul (talk) 13:34, 21 June 2010 (UTC)

4 Pi as in IBM System/4 Pi. The E3AWACS article the original commenter cited says "The signal and data processing is carried out on a high-speed powerful IBM 4PiCC-1 computer." A NASA page cited by the IBM System/4 Pi article says:

The AP-101S avionics box is 19.55 inches long, 7.62 inches high and 10.2 inches wide, the same as one of the two previous GPC avionics boxes. Each of the five upgraded GPCs aboard the orbiter weighs 64 pounds, in comparison to 114 pounds for the two units of the older GPCs. This change reduces the weight of the orbiter's avionics by approximately 300 pounds and frees a volume of approximately 4.35 cubic feet in the orbiter avionics bays. The older GPCs require 650 watts of electrical power versus 550 watts for the upgraded units.

which is a bit smaller, I think, than a System/360 Model 65. Guy Harris (talk) 07:32, 9 November 2010 (UTC)

There is apparently a Model 30 at the Catalunya Science & Technology Museum in Terrassa [1]

The 360 at Auckland's MoTaT is a Model 40, not a 30 (though 30 appears in some descriptions.) [2] Lawrencejw (talk) 12:44, 11 April 2012 (UTC)

Proposed new section: Component Names

I don't think we are going to reach any agreement about whether the IBM 2361 should be called a “peripheral”. How about a new section, just before the Peripheral Devices section, called Component Names. It would start with something like the following:

IBM created a new naming system for the new components created for System/360, although well-known old names, like 1403, were retained. In this new naming system, components were given four-digit numbers, starting with 2. The second digit described the category, as follows:

followed by the list currently in the Peripherals section. Sound reasonable? —Preceding unsigned comment added by John Sauter (talk • contribs) 19:49, 28 June 2010 (UTC)

Sounds good, except that a components section should probably include boxes not currently listed, e.g., 1052-7, 2250.

I have separated the sections, as described above. I have included the 1052; the 2250 was already on the list. Feel free to add to the examples, particularly components that already have Wikipedia entries. John Sauter (talk) 14:36, 29 June 2010 (UTC)

BTW, User:Chatul/References includes a number of hardware manuals formatted with {{cite manual}}, and I have a few more on my bookshelf that I can add if you need them. I've tried to include links to bitsavers where they exist, except for manuals new enough to still have copies on an IBM web site. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:36, 28 June 2010 (UTC)

If you have the System/360 model 50 Functional Characteristics manual, that would be a good addition to the otherwise fairly complete set of Functional Characteristics manuals at bitsaver. If you could scan it and contribute it to bitsaver, that would be most convenient. John Sauter (talk) 14:36, 29 June 2010 (UTC)

I used to have it, but I'm pretty sure that I got rid of it. Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:38, 30 June 2010 (UTC)

---

Hate to bring up a controversial subject but the 236x are not DASD! I'm going to make the Component bullet into two lines pointing out that these are "storage" and not DASD. Tom94022 (talk) 16:45, 11 April 2012 (UTC)

There's nothing wrong with bringing up a controversial subject. I question your formatting of the text: breaking the line in two looks ugly when viewing the article using narrow columns. I think it would look better without the line break. More subtantively, are you sure the IBM 2361 does not qualify as a direct access storage device? The medium (core) is accessed directly rather than sequentially (like tape or cards) and it is definitely storage. John Sauter (talk) 04:12, 12 April 2012 (UTC)

I think it is ugly either way; maybe I will play with a table, but that is a lot of work. Actually, DASD are not addressed directly but instead thru channel CCWs. Main Storage, sometimes called random access storage is addressed directly. AFAIK, the storage boxes were addressed as memory and not as DASD, that is they used a memory addresses and not DASD CCW's. In my view that makes them main storage, I could be wrong. Tom94022 (talk) 18:14, 12 April 2012 (UTC)

Well, DASD is just an acronym for Direct Access Storage Device. You are correct that the IBM 2361 was addressed as memory rather than with a channel, but I'm not sure that excludes it from being a DASD. IBM considered the 2361 to be an “extension” of main (processor) storage—see the references in IBM 2361. John Sauter (talk) 12:57, 13 April 2012 (UTC)

So by your definition IBM disk drives which are indirectly addressed thru the channel are not direct access devices? The System 360 Summary includes the 2361 under "Main Storage Features (p3-1, 17/62) and does not list it in "Direct Access Devices" in "Section 7, Input/Output Devices" (p7-1, 53/62). The 2365 is described as "Main Storage" for the M65 (p6-11, 41/62); it is also not listed in "Direct Access Devices". I think it is pretty clear that DASD and Main Storage are two distinct classifications that IBM unfortunately included in the 23xx number series. Tom94022 (talk) 16:22, 13 April 2012 (UTC)

I turned the list into a table which should work on all width screens Tom94022 (talk) 23:46, 13 April 2012 (UTC)

I think the term “direct access” was intended to be understood in contrast to “sequential access” which included cards, magnetic tape and paper tape—you could write a block without invalidating all later blocks, and you didn't need a lengthy rewind operation to get back to the beginning of a data set. Even stronger was the term “random access”, which meant that access to any byte of storage took the same length of time, no matter where the previous access was. (This was only approximately true for the IBM 2361 due to its interleaving combined with it releasing the processor after 3.6 microseconds but being busy for the full 8.0 microseconds.) Both direct access and random access components were included in the 23xx component names.

I think your new table looks quite good. 08:32, 14 April 2012 (UTC)

Thanks for the complement.

I am pretty sure "direct access" was coined to distinguish over "random access". Pre-S/360 IBM described memory as random access or sequential access but as they moved from drums, williams tubes, etc., into core they realized that random access meant the time to access any memory location was essentially constant (e.g. 2361) and independent of the current memory access location. Since disk drives did not fit that definition they invented "direct access" as in between random and sequential. Note the first disk drive was the RAMAC (Random Access ....)! During the 1960s IBM San Jose hosted a series of internal "RAMP" corporate meetings on disk drives and the like - Random Access Memory Products. By the late 60s the RAMP Product Manager became the DASD Product Manager. Tom94022 (talk) 22:30, 14 April 2012 (UTC)

Selector subchannels

My recollection is that selector subchannels existed only in the 2870, which was used by the larger models, not in the integrated channels of the /30, /40 and /50. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:51, 10 August 2012 (UTC)

I believe the multiplexor channel on models 30-50 could run a single device in “burst mode” but this is not the same as a selector subchannel. The relavent text from the System/360 model 65 functional characteristics manual is: “The multiplexer channel provides up to 196 subchannels, including four selector subchannels. The basic multiplexer channel has 192 subchannels; it can attach eight control units and can address 192 I/O devices. The basic multiplexer channel can overlap the operation of several I/O devices in multiplex mode or operate a single device in burst mode. One to four selector subchannels are optional with a 2870. Each selector subchannel can operate one I/O device concurrently with the basic multiplexor channel. Each selector subchannel permits attachment of eight control units for devices having a data rate not exceeding 180 kb. Regardless of the number of control units attached, a maximum of 16 I/O devices can be attached to a selector subchannel.” John Sauter (talk) 14:35, 12 August 2012 (UTC)

Also, note that the selector subchannels of the IBM 2870 each had their own bus and tag interfaces. This may be obvious to some, but it wasn't obvious to me until I read up on it. John Sauter (talk) 14:48, 29 August 2012 (UTC)

{{main}} tag in Direct access storage devices (DASD) section

The section Direct access storage devices (DASD) has a {{main|History_of_IBM_magnetic_disk_drives#IBM_S.2F360_and_other_IBM_mainframe_HDDs}} tag, but the 2301, 2303 and 2321 are not hard disk drives and thus are not discussed in that article. Shmuel (Seymour J.) Metz Username:Chatul (talk) 23:26, 26 August 2012 (UTC)

Agreed but so what, since the main form of S/360 DASD was HDDs and the tag reads. "See also". This whole section needs to be rewritten to focus on DASD which I will do so when I have some time. The place to start is with reference to IBM System/360 Component Descriptions - 2841 Storage Control Unit, 2302 Disk, 2311 Disk, 2321 Data Cell, 7320 Drum which I believe describes the first DASD on S/360 and probably represents the introduction of the term DASD to the art. Most of what is in the section should be links to parts of History_of_IBM_magnetic_disk_drives#IBM_S.2F360_and_other_IBM_mainframe_HDDs Tom94022 (talk) 17:19, 17 September 2012 (UTC)

Template:Main states

This template is not to be used as a substitute for inline links or as a "see also". Its usage should be restricted to the purpose described hereinbefore.

I corrected a typo in your reference. Shmuel (Seymour J.) Metz Username:Chatul (talk) 12:49, 19 September 2012 (UTC)

So why not just change it to Template:See_also and be done with it? Tom94022 (talk) 02:32, 20 September 2012 (UTC)

Bus and tag cables and connectors

The text states "Control units were connected to the channels with gray "bus and tag" cable pairs" yet the picture is of blue cables. I think these may even be post-370 as they are a lot thinner than I remember them being. As well as being inaccurate it is contradictory. Also, the bus and tag terminators are 370 and not 360. The 360 terminators were rectangular and not castellated. I will search for some more accurate photos and replace them - if I can (new to this!).Mark Triggers (talk) 13:02, 7 May 2011 (UTC)
OK - finally got around to uploading a picture of the correct 360-type terminators. I have used it to replace the picture of the later 370-type terminators. Now just need to get a picture of some grey (gray) channel cables. Mark Triggers (talk) 18:47, 28 December 2012 (UTC)

There is a pretty good set of photos here. I've asked them for permission to make a collage for Wikipedia and will post it if and when I get permission. The cables currently referenced in this article are not what I recall; 360 bus and tag cables I recall had one not two cables between the connectors as shown here. Tom94022 (talk) 21:36, 28 December 2012 (UTC)

Tom – the cables in your link are not original 360 cables. They are closer than the current picture (which have two thin cables per connector) but they are still blue. They may be functionally the same, but I’m pretty sure that 360 cables from the 60s were grey in colour. Mark Triggers (talk) 00:53, 29 December 2012 (UTC)

Here are a couple of my own photos that clearly show all the cables are grey photo1 photo2. Mark Triggers (talk) 01:41, 29 December 2012 (UTC)

While I can't recall any color other than grey, the color is not functional so it maybe they are third party cables or maybe IBM did make blue ones. The particular cable pictured in my link is purported to be for a 3480 which is S/370 from the 1970s. My recollection is there were no difference between S/360 and S/370 cables. The OEM manual gives the connector colors as light and dark grey but doesn't mention the cable color. If I get permission to use the photos, I suggest having one cable between connectors is reason enough to change until we can find a grey one. In the mean time I will keep looking for grey (or maybe I will Photoshop the one we have into grey :-) Tom94022 (talk) 03:40, 29 December 2012 (UTC)

The blue ones are a later higher speed version, according to "A SHAREd History of the Mainframe" the 3033 doubled the maximum channel speed to 3 MB/sec (the new cables were blue instead of grey but still had dark grey and light grey ends). Maybe I will Photoshop 04:09, 29 December 2012 (UTC)

Turns out it was easy to change color. How does this look? Any suggestions for editing. I might label the connectors "A connector" and "B connector" Tom94022 (talk) 06:57, 29 December 2012 (UTC)

Very good - far better than the blue ones. Labeling the connectors would be useful. Check your upper/lower case consistency "Cable" v "cable". Mark Triggers (talk) 13:50, 29 December 2012 (UTC)

Final version uploaded to same site; awaiting permissions before posting to Wikipedia commons. Again suggestions would be appreciated. Tom94022 (talk) 04:09, 30 December 2012 (UTC)

Final Final version uploaded; now have permission to use photos. Any suggested changes before I upload to WikiMedia? Tom94022 (talk) 03:22, 3 January 2013 (UTC)

To clarify, a S/360 channel had either two cables or three. A normal channel had one bus and one tage cable, while the 2-byte channel used for the 2305 Model 1 had two bus cables and one tage cable.

Agree, but I think the two byte wide channel was of such limited use in S/360 as to be TMI and relegated to a footnote at best. Tom94022 (talk) 23:16, 4 January 2013 (UTC)

There was definitely color coding of various versions of the B&T cables, but I don't recall the details. It's likely that the cables for data streaming required tighter tolerances than those for the older interlock protocol. Shmuel (Seymour J.) Metz Username:Chatul (talk) 21:45, 4 January 2013 (UTC)

I am pretty sure it was grey for standard channel and blue for the data streaming channel. As near as I can tell it was first introduced at 3 MB/sec in 1980 with MVS/SP1 and the 3380 attached to the 303x. Later it was raised to 4.5 MB/sec but I don't know if the cable color changed, I don't think so. Tom94022 (talk) 23:16, 4 January 2013 (UTC)ibm bm

Units

To answer the question in a recent edit, IBM used decimal units for I/O capacities and speeds. The only places in the S/360 literature where they misuse SI units, e.g., K, M, G, T, is in memory (RAM and ROM) capacities. Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:54, 3 October 2013 (UTC)

APL Shared Variables history and nomenclature

IBM released APL Shared Variables for the System/370, not for the System/360. While I have seen references to it as "APL SV", "APL-SV", "APL/SV" and APL\SV", I have not seen anything that gives an official abbreviation. The term "APL.SV" is used in the 5100 manuals, but not in any S/360 or S/370 manuals that I have seen. Hence I question the last edit, although something like "S/370 APL Shared Variables (APL.SV on the 5100)" might be appropriate. Shmuel (Seymour J.) Metz Username:Chatul (talk) 12:51, 17 October 2010 (UTC)

APL has nothing to do with System/370 (nor its successors). It is a software system (that happened to hun IBM mainframes). — Preceding unsigned comment added by Rbakels (talk • contribs) 21:05, 20 November 2013 (UTC)

Typical main storage size

IMHO the numbers quoted in the article are way too large. 32K and 64K machines were more common, especially for the lower processor capacity machines. Core memory was extremely expensive, millions of dollars per megabyte. Rbakels (talk) 20:45, 20 November 2013 (UTC)

If your opinion can be supported by references, you should correct the article. John Sauter (talk) 14:40, 23 November 2013 (UTC)

Were the S/360 models 60 and 62 ever sold?

The article includes the following paragraph:

The initial announcement in 1964 included Models 30, 40, 50, 60, 62, and 70. The first three were low- to middle-range systems aimed at the IBM 1400 series market. All three were sold first during mid-1965. The last three, intended to replace the 7000 series machines, were never sold and were replaced by the 65 and 75, which was first delivered during November 1965, and January 1966, respectively.

The following statement from page 11 of the Computer History Museum's Oral History interview of Frank Belvin (http://archive.computerhistory.org/resources/access/text/2013/03/102702246-05-01-acc.pdf) seems to contradict this. Belvin was describing his time at MIT's Lincoln Laboratory.

The first [S/360] machine we got was a standard 60 or 62 running OS/360 because we weren't planning to make a complete upheaval in the system.

-Jeff Ogden (W163) (talk) 00:32, 20 October 2014 (UTC)

It will shortly say they were never shipped according to Padegs, "System/360 and Beyond," IBM JRD, Sept 1981. Tom94022 (talk) 20:39, 20 October 2014 (UTC)

AFAIK, none of the model 60, 62, 64, 66, 70 or 90 were sold, and the 360/92 was relabled as 360/91 when it failed to meet performance goals. Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:25, 21 October 2014 (UTC)

Reading over the transcript, I think IBM must have proposed a model 64 or 66, but delivered a model 67. Frank says they ran OS/360, which means the model 67 was used as a model 65. We did the same thing at Stanford. Note page 12 where Frank says that IBM attempted to IPL TSS—they wouldn't have tried that with any model other than a 67. Also, on page 13 he says they had a model 67. John Sauter (talk) 10:08, 23 October 2014 (UTC)

Model 20

I believe the Model 20 was introduced later than 1966, perhaps in 1969 or 1970. I joined the company in the then Data Processing Division in 1968 and do not recall the M20 in the line during my training. It should also be noted that the M20 used an entirely different punch card - smaller than the standard, and with a new key punch that made small round holes rather than the Old Hollerith styleAbba1943 (talk)abba1943, 20"20 26 July 2008

This is all pretty well documented. See IBM archives for specifics, e.g. at ibm.com mainframe timeline. The Model 20 was announced November 18 1964, and first shipped in April 1966. It is referenced at various places in Pugh et al. (see article for citation), e.g. p. 445 which states: "The IBM 2560 Multi-Function Card Machine (MFCM) [was] announced in November 1964 with the System/360 Model 20." (We had a different explanation for the "MFCM" acronym, by the way, where the "CM" stood for "Card Muncher" due to the device's ability to mangle punchcards, and...I don't remember what the "MF" stood for.)

Note that, despite its name, the Model 20 was not really a S/360; thus Padegs (in "System/360 and Beyond", which is referenced in Pugh; see IBM System/370 for a full citation), p. 379, says "Model 20, although nominally called part of the System/360 family, was incompatible with System/360."

Note also that the S/360 Model 22 was announced in 1971, and the Model 25 was announced in 1968. Perhaps the above comment was related to memories of these other systems. Spinality (talk) 01:29, 28 July 2008 (UTC)

Abba1943 might be thinking of the IBM System/3, which did use a smaller 96-column punched card with round holes, and which, at least according to the Wikipedia article, came out in 1969. Guy Harris (talk) 23:42, 21 January 2010 (UTC)

Indeed. When I joined IBM in 1974, IBM System/3 with 96-column (96 character) punched cards was the big fashion. IIRC System 360 Model 20 is older, and differed in that it used 16 rather than 24-bit addresses, limiting the address space to 64KB (theoretically, because a 64 KB memory was "astronomic" at the time). The (fully incompatible) System/3 was probably developed by IBM in preparation of the threat that IBM would be split for antitrust reasons, and the establishment of IBM's "General Systems" division. — Preceding unsigned comment added by Rbakels (talk • contribs) 21:03, 20 November 2013 (UTC)

64KiB may have been astronomical for entry level systems at the time the 360/20 was announced, but it was inadequate for medium and large systems, and competitors had systems with more than 64 Ki words. By 1974, the low end 370/145 wasn't even available with a memory as small as 64 KiB. The S/3 was aimed at a different market, and didn't have to de4al with large workloads. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:37, 27 October 2014 (UTC)

Table of models

I think a table of 360 models would be helpful. I thought I'd start it here and it can be moved to the article when more populated. This is a first cut, assistance welcome. --agr (talk) 16:31, 8 October 2014 (UTC)

Proposed Table

Note: This table has been copied to the article, so any further edits should be made there.--agr (talk) 19:54, 26 October 2014 (UTC)

Model	Announced	Shipped	Scientific Performance (kIPS)	Commercial Performance (kIPS)	CPU Bandwidth (MB/sec)	Memory Bandwidth (MB/sec)	Memory (KiB)	Notes
30	Apr 1964	Jun 1965	10.2	29	1.3	0.7	8-64
40	Apr 1964	Apr 1965	40	75	3.2	0.8	16-256
50	Apr 1964	Aug 1965	133	169	8.0	2.0	64-512	Supported IBM 2361 Large Capacity Storage (LCS)
60 - 62	Apr 1964	never						Replaced by Model 65
70	Apr 1964	never						Replaced by Model 75
20	Nov 1964	Mar 1966	2.0	2.6			4-32	Low end, limited instruction set
91	Nov 1964	Oct 1967	1,900	1,800	133	164	1,024-4,096
64 - 66	Apr 1965	never						Replaced by Model 67
65	Apr 1965	Nov 1965	563	567	40	21	128-1,024	Supported LCS
75	Apr 1965	Jan 1966	940	670	41	43	256-1,024	Supported LCS
67	Aug 1965	May 1966			40	21	512-2,048	Dynamic address translation for time sharing
44	Aug 1965	Sep 1966	118	185	16	4.0	32-256	Specialized for scientific computing
95	special order	Feb 1968	3,800 est.	3,600 est.	133	711	5,220	Performance estimated as 2x Model 91 per Pugh p.394
25	Jan 1968	Oct 1968	9.7	25	1.1	2.2	16-48
85	Jan 1968	Dec 1969	3,245	3,418	100	67	512-4,096	16-32 KiB cache memory, extended-precision floating point
195	Aug 1969	Mar 1971	10,000 est.	10,000 est.	148	169	1,024-4,096	32 KiB IC cache memory. Performance estimated as 3x Model 85 per Pugh p.422.
22	Apr 1971	Jun 1971			1.3	0.7	24-32	A re-manufactured Model 30

Style Issues

Good idea, I have a copy of Pugh's "IBM's 360 and Early 370 Systems" which is a reliable source for the dates which I have changed above. The question is what do we mean by speed and can we quantify it in a way that is comparable across the models with one or a range of simple numberss? I'll see what I can dig out of Pugh. Tom94022 (talk) 18:31, 8 October 2014 (UTC)

Great, thanks. I'll be off for a few days, so feel free to move to the article if you think its ready.--agr (talk) 21:18, 8 October 2014 (UTC)

I'm making this into a sortable Wikitable which means the dates will be in the form of YYYY MONTH. I've started into this but it will take some time. I doubt if it will be ready by the time u get back - enjoy the time off Tom94022 (talk) 22:56, 8 October 2014 (UTC)

I changed the "Shipped" column, but note that neither YYYY MMMM nor MMMM YYYY columns currently sort correctly. Guy Harris (talk) 00:14, 9 October 2014 (UTC)

Thanks, they do sort correctly for other than the undated items - I'll try to figure a work around, maybe 9999 with a footnote for the never shipped. Tom94022 (talk) 00:39, 9 October 2014 (UTC)

Changing the "60, 62" to "60 - 62" fixed that sort :-) Tom94022 (talk) 00:41, 9 October 2014 (UTC)

The date columns do not sort at all for me (Safari Version 6.2 (8537.85.10.17.1) on OS X 10.8.5). I can click on the arrows and nothing happens. Do they sort for you and, if so, what version of what browser are you using? Guy Harris (talk) 05:26, 9 October 2014 (UTC)

It works for me with Chrome on WinXP and Win7 - I suspect yr problem is with Java and Safari. I'll check it out later on FF and UE Tom94022 (talk) 06:01, 9 October 2014 (UTC)

Using data-sort-value="" in the non-date cells I now have the first three columns sorting correctly in Chrome on WinXP and Win7. I also added the reference to Pugh for the two date columns. I'm not sure adding the day of the month to some cells adds much value and there would be a fair bit of research to find days for all of the announcements so i suggest going with Pugh as one reliable source with the two columns in the same format (YYYY MONTH) makes sense and will look better. Its probably impossible to find the day of shipment. But if someone wants to find the day of each announcement and make them sort properly I guess that's OK too. Tom94022 (talk) 06:21, 9 October 2014 (UTC)

Presumably by "there would be a fair bit of research to find days for all of the announcements" you mean that the IBM Archive "System/360 Dates and characteristics" page is not a reliable source, as that has a complete list of days for all the announcements. Guy Harris (talk) 07:39, 9 October 2014 (UTC)

The IBM Archive and Pugh are reliable sources and they both rely upon Padegs (the lack of a day for the 91 in the original edit confused me). I suggest linking to a book (or Padegs) is preferred over a web page. I do think having the same format in the two columns looks and reads better. Sorting doesn't seem to be working well for dates so it maybe better to go to a MMMMM YYYY format if date sorts are shown to not be reliable. I'm looking into sorting issues. Tom94022 (talk) 16:15, 9 October 2014 (UTC)

It's a bit of work but I can make date columns sort in order regardless of how they are displayed (on Windows XP or 7, with Chrome, FF and IE so far). So does anyone have any strong feelings about the format. Personally I would use "MMM YYYY" so that each entry is the same size which is sort of in accordance with WP:YR and I do suggest consistent format between the two columns which rules out "Month Day, Year" but I can go with any consensus. My test case is here with the Announced column fixed and the Shipped column unchanged; I'd appreciate feedback from Opera and Safari users. Tom94022 (talk) 21:38, 10 October 2014 (UTC)

I moved the working table here. Unfortunately John's subsequent edit to add the exact day to the sort field and use a letter to sequence same dates broke the sorting (the code stops sorting at any non-number).

• I suppose replacing it with a decimal digit might work but does anyone care how we sort?
• Does anyone have any preference for a date display other than the current Mon YYYY? Tom94022 (talk) 17:33, 11 October 2014 (UTC)

I have made the announce date sort keys all seven digits long, and all unique, so sorting by announce date should now be correct. Sorry for the breakage. John Sauter (talk) 22:19, 11 October 2014 (UTC)

How are CPU bandwidth and memory bandwidth defined? I suppose these terms are defined in the reference, but the reference isn't on the web. John Sauter (talk) 10:17, 23 October 2014 (UTC)

KB versus KiB

Memory sizes - Wikipedia:Manual of Style/Dates and numbers "The IEC prefixes kibi-, mebi-, gibi-, etc. (symbols Ki, Mi, Gi, etc.) are rarely used, even in technical articles, so are generally not to be used". Peter Flass (talk) 16:04, 13 October 2014 (UTC)

I read the [Manual of Style section on Quantities of bytes and bits], including its footnote. It recommends that an article adopt a consistent style for using or not using the binary prefixes. I am personally uncomfortable continuing the ambiguity inherit in using the same abbreviations for different values. If we decide not to use the binary abbreviations, I advocate that we not use any, and spell out the number of bytes. Thus, instead of writing 5,220 KiB we would write 5,345,280 bytes. Note that the recommendation in the Manual of Style was last discussed in 2008, and there was significant dissent. John Sauter (talk) 09:29, 14 October 2014 (UTC)

Content Issues

The model 20 submodel 5 supported 32 KiB of core memory. The 1967 functional characteristics manual doesn't include the submodel 5, but the 1969 DPS manual does. Also, I advocate omitting the models that were announced but never shipped. In addition to the 60, 62 and 70, there were also the 64 and 66, which were replaced by the 67. John Sauter (talk) 03:20, 9 October 2014 (UTC)

John does bring up the difficult subject of extensions to various system models. For example I seem to recall the M20 was actually "enhanced" at four or five different times, the second adding tape and the last being the submodel 5. Not sure if this applies to the other models nor am I sure what to do in the table. Perhaps ignore the extensions leaving them for the article?

I'm also not sure what to do about the never shipped models - I could live with John's suggestion to leave them out but see no harm in leaving them in. Tom94022 (talk) 06:01, 9 October 2014 (UTC)

I would follow the precedent of IBM's page http://www-03.ibm.com/ibm/history/exhibits/mainframe/mainframe_FS360.html, which lists the 60, 62, and 70 but none of the other models that never shipped. The 60, 62, and 70 are important because they were part of the original announcement. There was a year's gap before their replacements were announced and another 7 months before delivery. As for the Model 20's submodels, I think they are best handled within that model's own article, maybe with a reference in this article's table.--agr (talk) 00:59, 14 October 2014 (UTC)

Reading the reference carefully, I think the model 95's memory was 1 MiB thin film plus 4 MiB core. I have corrected the memory size to 5,220 KiB. John Sauter (talk) 09:09, 14 October 2014 (UTC)

Some of the numbers under memory bandwidth look bogus. The model 195 appears to have much slower memory than the model 95, but its performance numbers are much higher. Also, how can the model 75 have higher memory bandwidth than the model 65, since they use the same memory boxes? John Sauter (talk) 10:21, 23 October 2014 (UTC)

I extracted Padegs Table 1 for the S/360 as a spreadsheet and placed it here so anyone can down load and check the calculations and content. I did use maximum numbers, that is maximum width/minimum cycle. As I interpreted the table, the 75 was a 4 byte transfer while the 65 was a 2 byte transfer. Again according to Padegs, the 95 had i MiB of fast semiconductor memory not present in other S/360s. Thoughts, comments and/or corrections appreciated. Tom94022 (talk) 18:48, 23 October 2014 (UTC)

The M65 functional characteristics manual says that "Each 2365 Processor Storage has a basic 750 ns storage cycle, with access to eight bytes (doubleword) in parallel." and that "The 2361 Core Storage is a large capacity direct access core storage unit. It has a basic 8-usec storage cycle, with access to two words (eight bytes) in parallel." The M75 functional characteristics manual says that it also uses the 2365 and 2361, but that the 2365 Model 3, as used by the M75, can be four-way interleaved, which might be the doubling you see in your interpretation of the table (which might be off in that "N byte transfer" should perhaps be "N word transfer").

A Google search for "360" "model 95" "thin film" found items speaking of the M95 having thin-film memory. A search for "360" "model 95" "semiconductor memory" found, among other things, a page saying that the semiconductor memory on the M95 was for "a 16-bit, system-protect memory array", so perhaps it was used for storage protect keys, with thin film used for main memory. (That page also says "With IBM’s first all-semiconductor memory machine, the System 370 Model 145 announced in October 1970, the company’s designs graduated to 64-bits for the cache memory and 128 bits for the main memory.") Guy Harris (talk) 19:31, 23 October 2014 (UTC)

My bad assumption the 95 high speed main memory was semiconductor - thanks for the research; Padegs shows 1 MiB of 180 ns "processor storage" which we now know was likely thin-film. According to Pages footnote 1, all other "processor storage" was core. Probably should footnote this. 18:36, 24 October 2014 (UTC)

The 195 had a cache; the 95 did not. Presumably the cache negated the diffence in memory speeds. Of course, the 195 also had more overlap within the CPU. Shmuel (Seymour J.) Metz Username:Chatul (talk) 16:56, 24 October 2014 (UTC)

As did the 85 per Padges, should note this also. Tom94022 (talk) 18:36, 24 October 2014 (UTC)

And per Pugh; it was semiconductor memory, but another hit for a search for "360" "model 85" "semiconductor memory" (a comp.arch thread) claims it wasn't ICs, but ASLT, saying the 195's cache was ICs. Guy Harris (talk) 19:25, 24 October 2014 (UTC)

To clarify, both the 85 and the 195 used a cache, both the 85 and 195 overlapped instruction fetching with instruction execution, but only the 195 had overlap within the execution unit. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:21, 27 October 2014 (UTC)

Thank you for posting the table from Padegs. The Maximum Bus Width and Cycle Time columns contain some misunderstandings of the models 65 and 75. According to the Functional Characteristics manual, referenced above, the model 65 had a memory bus width of 64 bits, which is 8 bytes. The cycle time of its memory was 750 nanoseconds, but with two-way interleaving the effective cycle time could be reduced to 400 nanoseconds. Thus the entries should be cycle time 400 to 750 nanoseconds and max bus width 8 bytes. The model 75 uses the same memory bus width and cycle time as the model 65, but adds four-way interleaving. More interleaving does not increase the memory bus width or change the cycle time—it just increases the likelyhood of getting a 400 nanosecond cycle rather than a 750 nanosecond cycle. Thus the entries for the model 75 should be the same as for the model 65.

The inclusion of cache memory in the later models makes the effective memory cycle time even more complex. We are getting into the area of Original Research and Too Much Information. I suggest we not include any memory bandwidth information—the commercial and scientific benchmarks should be enough. John Sauter (talk) 15:33, 25 October 2014 (UTC)

I would advise against replacing actually cycle times with assumed effective cycle times. Show benchmarks, the actual access time, the actual cycle time, the interleaving and the cache characteristics, and let it go at that. If you must provide an effective cycle time, show it as a separate column and explain how it is calculated. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:48, 27 October 2014 (UTC)

Speed

Is Computer Speeds From Instruction Mixes - pre-1960 to 1971 an acceptable reliable source for this article? If so I propose using the "ADP Mix" data and estimating the missing 360/195 from data in Pugh on the M65 - M91. comments? Tom94022 (talk) 17:59, 11 October 2014 (UTC)

I think the reference is sufficiently authorative, or at least we are unlikely to find a better one. I suggest we use both the Gibson and the ADP columns, labeling them "scientific" and "commercial" processing speeds. We can omit the speed of the model 195, unless we find an authorative source for it. John Sauter (talk) 22:26, 11 October 2014 (UTC)

I think we have a reliable source in Pugh that "the 195's internal performance averaged perhaps three times that of the fastest Model 85" so we would not be violating WP:SYN if we imputed a performance.

Phister, "Data Processing Technology and Economics," Table II.2.11.1 System Characteristics, p. 342-5, is another source for performance for M20 thru M65 to which we could impute M75-M91 performance from Pugh Table 7.1 which gives performance as ratio's of M65, again without violating WP:SYN.

I'll put up several possible measures that we can look at and see which if any belong in the table we put in the article. Tom94022 (talk) 20:42, 12 October 2014 (UTC)

I've now looked at several performance calculations and measurements for S/360 and think Longbottom is as good as any and reliable. If this is OK with most, then we might consider normalizing about one processor, perhaps the M65 which seems to be common to most data sets I've looked at. I charted the data sets and if I have a chance I might put them on Wikimedia. I'd like to see a consensus here before posting the table since the last thing we need is an edit war over whose calculations or measurements are best. Tom94022 (talk) 23:29, 13 October 2014 (UTC)

I don't see any need to normalize the data. I think quoting the numbers in Longbottom, plus estimates with good references, is fine. Nevertheless, I would like to see Tom's chart. John Sauter (talk) 10:00, 14 October 2014 (UTC)

I like Longbottom as a source, but there is still the question of which speed value to report, Gibson mix vs the UK Gov's ADP mix. Gibson, intended for scientific applications gives a .129 weight for floating point operations, while ADP is integer only, and therefore more comparable to the Dhrystone mix used in our Instructions per second article's speed table. The ADP speed is more in keeping with what the low end machines were used for. The difference is small at the high end but can be dramatic at the low end, almost a factor of three for the model 30. Also note that Longbottom at http://www.roylongbottom.org.uk/whetstone.htm has Whetstone benchmarks for both the 360/65 and the 360/195. The values are compiler dependent, but as long as the same compilers are used to calculate a speed ratio, that should be good enough for our table. I'd prefer using this to Pugh's "perhaps three times that of the fastest Model 85".--agr (talk) 17:23, 14 October 2014 (UTC)

I believe that it is important to use consistent instruction mixes when comparing speeds, and that the article should identify the instruction mix(es) used, and the compiler version and release if the benchmark is not written in assembly language. If consistent data are not available then the article should so state. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:43, 14 October 2014 (UTC)

Please take a look at the Longbottom reference http://www.roylongbottom.org.uk/cpumix.htm which explains the instruction mixes used. We could create an article on the Gibson Mix and the ADP mix that lists their instruction weightings, and reference that. The only place where a compiler would be involved is in comparing the 195 with the 65 using the data at http://www.roylongbottom.org.uk/whetstone.htm. The compilers used are specified in that source so it is easy to get benchmarks based on the same IBM compiler. --agr (talk) 21:12, 14 October 2014 (UTC)

I've posted a comparison of Longbottom, Knight, Pugh and Roberts here. I think Longbottom extended by Pugh is reasonable. I'm not sure about using Longbottom Whetstones to impute M195 - his ratio is 9:1 and I expect Pugh's 3:1 has a lot more underlying data than the few samples of Longbottom. Tom94022 (talk) 01:03, 15 October 2014 (UTC)

BTW, the distinction between scientific and commercial computing is pretty much moot today, so in the interest of keeping the chart small I suggest we just go with the scientific column and explain the selection in the footnote. I have the Padegs article which has a table of processor speeds and feeds but no overall performance. I could put up something like processor bandwidth (data flow width/cycle time) or memory bandwidth (bus width/cycle time) - if I have time maybe I'll post both into the prototype chart so we can see what those numbers look like. Tom94022 (talk) 16:30, 15 October 2014 (UTC)

Why is it moot? Given the significant differences shown in the table it doesn't seem to be moot when comparing the various models. Peter Flass (talk) 17:36, 15 October 2014 (UTC)

Longbottom's Whetstones are reporting actual measured speeds using specific compilers, but we only have them for the 65, 67 and the 195. For the 65, where we have all three numbers, the Gibson and ADP are almost identical (563 kips vs 567), while the Whetstone Fortran opt2 single precision is very close at 521 kips, a ratio of 0.93. The Longbottom Whetstone 9:1 ratio Tom mentions I suspect is for the 195 vs the 65 (I get 9.25 for Fortran op2, single precision, 11.33 for double precision). The Pugh's 3:1 is comparing the 195 with the 85, not the 65. If we use the Gibson mix, the 85 is 5.76 times as fast as the 65. That makes the 195 to 85 ratio 1.605 for single precision, 1.966 for double precision, which is not that far off from Pugh, but data based.

If we are to just use one number for speed in the table, the Longbottom commercial-oriented ADP values would be best. The are very close to the scientific-oriented Gibson mix on the bigger models (65 and up). The smaller models did not have floating point hardware, so they are severely penalized by the Gibson mix. But applications for the smaller 360s generally did not involve floating point calculations, so the ADP mix provides a much better comparison for overall performance. So I would just report the Longbottom ADP kips for all machines except the 195, where I would use the highest Wheatstone Longbottom reports, with a footnote mentioning the fact.--agr (talk) 19:22, 15 October 2014 (UTC)

The smaller models most definitely did have floating point, and every 360/40 and 360/50 that I have used was so equipped. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:32, 15 October 2014 (UTC)

FP was optional on the /30, although I didn't know anyone who had it. Peter Flass (talk) 21:10, 15 October 2014 (UTC)

My recollection is that the 50 and maybe the 40 were considered mid-range machines and this is reflected in smaller ratios of their Gibson mix to the ADP mix (no f.p.) speeds (1.57 and 1.27 respectively, vs 2.8 for the model 30 and close to one for the 65 and larger models. However, these difference just underline the reason to use the ADP mix numbers. I wouldn't object to giving both speed numbers, but using just the Gibson mix would greatly understate the utility of the sub-65 models in their most common uses.--agr (talk) 21:09, 15 October 2014 (UTC)

I like Tom's chart, except for the horizontal axis. IBM's marketing numbers are good for identification, but not for plotting. Think how different the linear fit line would have been if IBM had chosen to call their model 195 the model 5 instead. How about using the purchase price, discounted based on date of first shipment? Calculating that would require some effort, but would produce a much more meaningful chart. Each spot could be marked with the model number. John Sauter (talk) 16:08, 16 October 2014 (UTC)

I put two measures of speed into the table and added them to the graph here. At this point I think the table maybe too complex for Wikipedia - should we pick one or at most two measures? The graph is just for context to help pick the measurement(s) and not intended for this article. Frankly I was surprised how closely things tracked, which suggests any one or two measures we agree upon is acceptable.

I personally have no preference between Longbottom Scientific or Commericial - do we need both?

I do think relative is better than absolute, particularly since we have Padegs saying the target at announcement was 25:1 and it grew to at least 200:1. BTW I did relative M65 in the graph since every source included it, but the table could go relative M30 to comport with Padegs.

I do think for consistency we should stick with Pugh for both the M95 and M195 estimates if we go with Longbottom.

Comments? Tom94022 (talk) 22:46, 19 October 2014 (UTC)

I don't think the table is too complex at all. Compare with the three specification tables at Mac mini, for example. I would keep both Scientific and Commercial, as that is how the market looked at performance back then. If anything, I would add the 1401 and 7094 for comparison purposes, and maybe a modern laptop. My one quibble is using Pugh for the 195 when we have reported measurements that suggest he over-estimates the 195/85 ratio. That said, I think it is time to put the table in the article. As for the graph, maybe one speed is sufficient, commercial giving a better end to end comparison for the 360 line.--agr (talk) 15:27, 21 October 2014 (UTC)

I can't understand why the Model 44 gets such a high rating on the commercial workload. Since all storage-to-storage instructuions had to be simulated, I would think it would rate quite low. Peter Flass (talk) 21:49, 26 October 2014 (UTC)

The 44's memory was twice as fast as the Model 50's, but the commercial performance score is only a little bit better. And remember these are calculate performance scores, not measured benchmarks.--agr (talk) 22:12, 26 October 2014 (UTC)

Cite errors

I tried copying the table into the article and ran into cite errors for the references in the table headings. It didn't work in the sandbox either. Any suggestions?--agr (talk) 13:46, 26 October 2014 (UTC)

Only genuine manufacturer-certified ASCII double-quote characters can be used around the "name" parameter to the <ref> tag, not those counterfeit "smart quotes". :-) I fixed the table. Guy Harris (talk) 18:56, 26 October 2014 (UTC)

Thanks, that solved the problem. I wonder why the smart quotes work in Talk space? I've now added the table to the article so any further edits should go there.--agr (talk) 19:52, 26 October 2014 (UTC)

I'm not sure they work in Talk space - I tried some test edits (not saved, just previewed) using smart quotes, and <ref name=”name” />, following an earlier reference with <ref name=”name”>, fails with a warning on IBM System/360 but fails silently, in that no reference tag appears, on Talk:IBM System/360. Perhaps the warnings don't display in Talk space for some reason? Guy Harris (talk) 20:37, 26 October 2014 (UTC)

And, in fact, that's what was happening in the smart-quoted version of the table when it was on the Talk page - there was no reference tag for the memory bandwidth column. Guy Harris (talk) 20:41, 26 October 2014 (UTC)

Better late than never, I guess. I thought the M20 deserved more information on its relative incompatibility with the other models. If people disagree, feel free to revert the edit. Peter Flass (talk) 21:43, 26 October 2014 (UTC)

Does this make sense?

The S/360 was the most expensive CPU project in history. (The most expensive project of the 1960s was the Apollo program for moon exploration. IBM's System/360 was the second most expensive. S/360 machines were also heavily used in the Apollo project.)

Does this make sense to you? --Spankthecrumpet 13:38, 4 Jun 2005 (UTC)

No, and I removed it. Feel free to rephrase it and put it back. Removed text follows below --Apoc2400 08:06, 26 October 2006 (UTC)

 
 ===The project's size and gravity===
 The S/360 was the most expensive CPU project in history. (The most expensive project of the 1960s was the Apollo program 
 for moon exploration. IBM's System/360 was the second most expensive. S/360 machines were also heavily used in the Apollo project.)
 Fortune Magazine at the time referred to the project as IBM's "$5 billion gamble," and they were right. IBM absolutely bet
 the company on the System/360 (US$5 billion in 1964 dollars translates to about $30 billion in 2005 dollars), which paid off.
 

Seems to me that it isn't easy to say what a "project" is. Different S/360 processors were built by different groups, following the described architecture, but implemented separately. OS/360 was a separate project, and then branched off to DOS/360, TOS/360, and BOS/360. Are each separate, or part of the overall project? Gah4 (talk) 02:18, 17 February 2015 (UTC)

Naming the System/360 family

There's a certain awkwardness in pages I've seen that talk about IBM mainframe systems, in that they sometimes refer to System/370, sometimes System/390, sometimes z/OS and so on.

As time goes by, pages will become out-of-date, and each rebranding will mean someone has to add a new brand name to existing pages.

I would suggest that something like "System/360 Family" be used as a generic name which doesn't need changing, and which could point to its own page on which the specific individual members of the family might be enumerated or linked.

I could just go ahead and do it, but I'd like to hear opinions first.

Brent Longborough 12:24 Dec 27, 2002 (UTC)

I think we should just merge all those pages into one, and then create redirects from all alternative names. — Monedula 07:58, 27 Aug 2004 (UTC)

I, too, have noticed this quite unsatisfactory condition. Another example is the IBM mainframe article, which is mostly about the S/360 family while also listing some previous systems. Please note the following, anyway: In the interest of Wikipedia standardization, the title of a potential single "S/360..." article should be "System/360 family" (lowercase "f"). --Wernher 18:59, 27 Aug 2004 (UTC)

Would "family" or "series" be more appropriate? e.g. IBM 700/7000 series and UNIVAC 1100/2200 series -- RTC 20:53, 29 Sep 2004 (UTC)

I would suggest 'family', because there were/are a wide range of models available at the same time, often quite different internally (designed by different laboratories in IBM), covering different uses. 'series' perhaps gives the sense of a line of machines, each one a follow-on to the previous model. mfc

Makes sense. -- RTC 23:36, 30 Sep 2004 (UTC)

Frankly, "System/360 family" is not an appropriate designation. While IBM always intended to provide backward compatibility (i.e. the capability of newer machiens to run older applications), technically the System/360 successors not only got different names for marketing reasons, but were truly different. Virtual memory was new for System/370 (with some minor exceptions), and 31-bit addressing was new for "XA", later "ESA" and System/390. I guess in common parlance all these machines are referred to as "IBM mainframes". (IBM had never another "mainframe"). — Preceding unsigned comment added by Rbakels (talk • contribs) 20:53, 20 November 2013 (UTC)

I like S/360 and successors. The later architectures are backward compatible for user code, but not for privileged instructions. But S/360, S/370, XA/370, ESA/370, ESA/390, and z/ are different, but related, families. Gah4 (talk) 02:20, 17 February 2015 (UTC)

1620 emulation?

The article state that the IBM 1620 was one of the architectures that the IBM 360 could emulate in microcode. I'm dubious about this claim. I don't remember such an option and I can't imagine much of a market for it then (the IBM 1130 was the successor in the 1620's market niche), but it's hard to prove a negative. Does anyone have a reference?--agr 19:32, 7 February 2006 (UTC)

I don't have a reference, but it seems reasonable. The 360 architecture was the first one to use microcode, wasn't it? Customer's existing base of applications was then, and remains today, one of the roadblocks to moving to a new architecture. I am old enough to remember when the IBMPC was first introduced. One of the points made, during the first months of introduction, was that there were tools to recompile your existing CP/M applications, so they would run under MS-DOS, preserving customer's software investment. The enduring legacy of backward compatibility with CP/m was a chain around MS-DOS's ankle for decades. -- Geo Swan 19:55, 7 February 2006 (UTC)

Microcode existed long before S/360; Maurice Wilkes came up with the idea, and it was used in an early research machine, and might have been used by processors before S/360. That doesn't mean that the 1620 was necessarily emulated in microcode by any S/360 machines, however; it might have been simulated in software, if that provided adequate performance (S/360 hardware could, at least, add two decimal digits without doing a table lookup, unlike the 1620 Model I :-)). I think pure-software simulators existed to simulate at least some other IBM computers on S/360. Guy Harris 20:29, 7 February 2006 (UTC)

I am old enough to remember when the IBM 1620 was first introduced. :-)>>> Some programmers at the time were unhappy that it was not compatible with the older IBM 650. IBM's main interest was back then leasing or selling hardware. The low end 360's were targeted at the commercial market and had rather poor scientific performance. It's unlikely many 1620 shops upgraded to these machines. The larger 360's were much more expensive than a 1620 so again there were likely to be few upgrades in that direction. The IBM 1130 was the preferred upgrade path for 1620 customers. And microcode was not cheap to develop or install. [3] All the literature I can find only mentions 1400 series emulation on the low end 360s and 7000 emulation on the high end machines.

As for software emulation, it may have existed, but I never heard of it for the 1620. The idea was known, but it would have been difficult to simulate a 1620 in the memory available on a low end 360/30 (64KB max, and that was an expensive configuration). And performance would have been pretty poor, even without the need for table look up. The main reason for writing in assembly language was performance, so simulation of such programs would have been less than ideal. By contrast most 1400 software was probably written in assembly language because memory space was tiny. Also I believe most 1620 software was written in Fortran, which provided a relatively easy upgrade path for most customers. --agr 23:30, 7 February 2006 (UTC)

See: converting to IBM System/360 page 5.

"Compatibility features and associated emulator programs are designed to protect your investment in 1620, 1400- and 7000-series programs. Emulators make it possible for you to execute your current programs on System/360 with little or no reprogramming. Using these features, the appropriate model of System/360 will normally execute your 1620,1400- or 7000-series programs as fast or faster than they run on your present system"

-- RTC 01:00, 20 September 2006 (UTC)

See: Laurie Robertson (Apr-Jun, 2005), "Anecdotes," IEEE Annals of the History of Computing, vol. 27, no. 2, pp. 82-84, IEEE {{citation}}: Check date values in: |date= (help)[4]

"A couple of years later, Fairchild bought its first IBM 360/30. It had firmware microcode stored in ROM that let it emulate both the 1620 and 1401. ..." (Emphasis added)

I found this by a Google search for ibm 1620 microcode, it was the first item. Apparently at least one 360/30 was delivered with 1620 emulation microcode. -- RTC 23:35, 27 September 2006 (UTC)

Thanks, I stand corrected. Nice find. I guess the moral is "Google first, ask questions later." --agr 23:46, 27 September 2006 (UTC)

It is possible that IBM sold a small number of systems with 1620 emulation, like the Fairchild one above, but when they saw the poor sales reevaluated the market and designed the 1130 to fill the void left below the low end 360 systems. A very similar problem occured when IBM attempted to sell the IBM 7070 as a common replacement for both the 650 and 705. It did so poorly as a 705 replacement that they were forced to design the IBM 7080. -- RTC 23:48, 27 September 2006 (UTC)

The 1620 simulator was a deck of cards that was booted. It was basically a stand alone program that ran on the bare 360 without running under an operating system. I believe it was compiled and punched out with the various options for its emulation compiled in. When I was in college, the big machine down the street was a 360/40, and they kept a copy of the simulator around (supposedly) in case of a catastrophic failure of one of the 1620s that were on campus (supposedly with options to match the features in the better optioned 1620). Also, in reading the model specific CPU manuals, I do not recall ever seeing an option listed for a 1620 emulator in firmware on any of the models. I expect the simulator probably could have squeezed into a 64K machine, if only emulating a 20k 1620, since that would have left about 45000 bytes for the simulator. I vaguely recall seeing the simulator run one time, but it was (of course) many years ago. Alan Larson 00:09, 17 October 2007 (UTC)

It would be worth adding monthly lease rates and purchase prices to the article, but i think we are talking about very different price points. The 360/30 was the low end of the 360 line, but still a medium priced machine. The 1130 was very cheap at the time and affordable for a small engineering company or a high school. The 360/30 was not considered a good scientific machine. Also most 1620s and 1130s were run open shop. i doubt many 360/30s were. Remember 360s had to be restarted to run in emulation mode and it tied up the machine(the 370s fixed that). It may be that some large customers had production programs running on the 1620 that they didn't want to convert, but I doubt there were many. As I mentioned above, converting a Fortran program was pretty easy. My guess is that the 1620 emulator was mostly used as a selling tool to get customers to upgrade to the more expensive 360 line.--agr 18:22, 28 September 2006 (UTC)

Thats also the reason the 360/20 was done (to fill the void left below the 360/30), but it was so stripped down that emulating much of anything on it would have been rather silly (even calling it a 360 was stretching things a bit!). And I suspect even the crippled 360/20 would have been overpriced for a lot of 1620 customers (e.g., 20,000 digit machines) which would be the only 1620s the 360/20 could hope to emulate in its tiny memory. A 1620 customer with a full up system (i.e. 60,000 digits, 4 - 1311 disks, 1622 card reader/punch, 1443 printer, 1627 plotter, and a lot of investment in custom in house special purpose code written in SPS) and outgrowing that system would likely be very willing to get a 360/30 with 1620 emulation, as a "transition" machine, then upgrade later to one of the larger machines. Also not all 1620s were used as scientific computers. For example several newspapers (e.g., The Kansas City Star) used the 1620 for typesetting with custom software all written in SPS. A 360/30 with 1620 emulation could easily handle typesetting for a newspaper with similar if not better performance to the old 1620. -- RTC 19:44, 28 September 2006 (UTC)

I'm sure it could, but I wonder if it would be cost effective (not that that ever stopped an IBM salesman). Not only would the lease rate be higher, but you'd probably need a bigger computer room, power and air conditioning. Did the 360/30 have a paper tape punch available? Might be cheaper to keep running the 1620. In any case, I don't think such situations represented the bulk of the 1620 business. Might be fun to contact the Kansas City Star to see if anyone remembers.--agr 21:41, 28 September 2006 (UTC)

Check the 1966 and 1968 history of Krohm International Ltd. It covers the Kansas City Star and their 1620. They upgraded to a 360 (not sure which model though, probably either a 30 or a 40, but it almost certainly had 1620 emulation). This was then replaced by a PDP-11. -- RTC 17:30, 29 September 2006 (UTC)

Very interesting, but I interpret what he said a little differently. It sounds to me like IBM was pushing an S/360-based typesetting product of its own and convinced the Star to upgrade to it. I suspect the IBM system was far more ambitious than the Star's the home-grown 1620 paper tape system. It likely had multiple terminals, perhaps IBM 2741s, for direct input of text instead of using paper tape produced off-line. I doubt the 1620 software was reused, tho the emulation capability may have been part of the sell. --agr 19:13, 29 September 2006 (UTC)

Regarding emulation, it seems that the later versions (submodel 5) of the 360/20 had 1401/1440 emulation microcode. Unlike earlier 360/20s, the submodel 5 kept microcode in core. That allowed one to IMPL (probably the term didn't yet exist) either the 360 or 1401/1440 microcode off cards. The microcode emulation options for each processor are described in its "Functional Characteristics" manual. Read through each one to see which are available. Software emulation, with no microcode assist, might be documented somewhere else. Gah4 (talk) 19:38, 17 February 2015 (UTC)

Type of tag-and-bus cable.

The recent enumeration of type of tag and bus cables is missing one; the double wide bu used for the 2305-1, containing on tag cable and two bus cables. Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:46, 12 June 2015 (UTC)

AFAIK, the second Bus cable was physically the same, just another IBM channel bus cable. Tom94022 (talk) 01:35, 13 June 2015 (UTC)

bus and tag cables

The article mentions the "gray "bus and tag" cable pairs" but the photo shows twin blue cables, which seem to something else or a later modification. Does anyone have a photo of the originals? Peter Flass (talk) 20:01, 15 February 2015 (UTC)

I think I have a photo and if so will post it in the commons and then here. The blue cables I think are later higher speed cables, same physical connectors. Tom94022 (talk) 20:51, 15 February 2015 (UTC)

Done Tom94022 (talk) 22:11, 15 February 2015 (UTC)

RAMAC 3 Direct Access Storage Device

The following sentence was added to the bus and tag section of the article recently: “For instance, the RAMAC 3 Direct Access Storage Device specifies that tan bus-and-tag cables only are to be used upstream, while both tan and blue cables were allowed downstream.” The sentence has no citation, and I am unable to locate any information about a RAMAC 3 Direct Access Storage Device. The IBM 350 was known as RAMAC, but it was not used on System/360. Does anyone know what device is being referred to here? I propose to delete this sentence unless a citation can be found or the device name corrected. John Sauter (talk) 14:58, 12 June 2015 (UTC)

Much to my amusement, IBM recycled the name RAMAC; I on't recall whether the RAMAC 3 supported bus&tag or whether it required ESCON.

A quick search in IBM RAMAC 3 Array Storage at page 103/226 indicates the RAMAC 3 used a DDC (Director Device Connection) cable between the array and its storage control unit. The SCU was either ESCON or parallel channel. Unless anyone has anything different to add, I propose we delete the edit. Tom94022 (talk) 01:35, 13 June 2015 (UTC)

Thanks for the reference. According to page 9, the RAMAC 3 supports only ESCON, not parallel channel. I will delete the sentence from the article. John Sauter (talk) 13:28, 13 June 2015 (UTC)

I'm confused. Tom's reference said that it did support parallel channels. Was the p.9 reference to the same device? Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:13, 15 June 2015 (UTC)

Tom's reference says, on document page 9:

Channel attachment is ESCON only; parallel channel support is not provided. The minimum channel configuration is 4 ESCON channels on a 9390-001 and the maximum is 32 on a 9390-002 (16 on each storage control).

If "page 103/226" means "PDF page 103 out of a total of 226 PDF pages", PDF page 103 is document page 89, and it says

Four new DDC cables are required to connect the RAMAC Storage Frame to the 9390 or 3990-6 storage control. The new cables are needed to support the higher data rate and are of a higher specification than the standard DDC cables. The maximum length of the cable is 200 ft. Current DDC cables cannot be used with RAMAC 3.

Document page 77 says:

The 9390-001 supports ESCON channels only. If you want to support RAMAC 3 Array drawers from a parallel channel host, you can attach a RAMAC 3 Storage Frame to a 3990-6, which supports parallel channels.

so it sounds as if you can attach RAMAC storage frames to a 3990-6, which supports parallel channels, or a 9390, which doesn't. I guess the 3990 was a pre-RAMAC storage control, which also support attaching the RAMAC, and the 9390 is a Shiny New storage control intended to work with the RAMAC. Guy Harris (talk) 18:08, 15 June 2015 (UTC)

The 3990 as the successor to the 3880. My recollection is that the 9390 was a disk drive or array with an imbedded controller. Shmuel (Seymour J.) Metz Username:Chatul (talk) 19:33, 16 June 2015 (UTC)

I apologize for missing the references to attaching the RAMAC 3 to a 3990-6, which does support parallel channels. However, the sentence still has no support, since the document does not say that attaching the RAMAC 3 to a 3990-6 places any requirements on the bus and tag cables. John Sauter (talk) 16:56, 17 June 2015 (UTC)