Talk:Mainframe computer/Archive 2

Capacity
How many terrabytes? Gazillion bytes?--Jondel 11:44, 21 October 2005 (UTC)
 * Three bajillion. RossPatterson 02:54, 14 November 2006 (UTC)

Reference for IBM mf
Could use a scale-reference for this picture 69.28.40.34 19:52, 2 October 2006 (UTC)
 * The z890 is about the size of a common US refrigerator. Or maybe a Sub-Zero, but not much bigger. RossPatterson 03:04, 14 November 2006 (UTC)
 * That doesnt help people outside the US very much 130.246.132.26 13:30, 29 November 2006 (UTC)

Other Mfg's
I just Wanted to point out that Hitachi Data Systems, HDS was one of the main players in this industry for a very long time, and even toppled IBM for a while. It would be good to mention them. —Preceding unsigned comment added by 24.20.186.59 (talk • contribs) 06:54, 13 July 2006
 * If by "very long time" you mean 15 out of 45 years, yeah, it was, along with Fujitsu. But "toppled IBM"?  Puhleeze.  RossPatterson 03:02, 14 November 2006 (UTC)
 * Added a Google translation of the Japanese Wikipedia article on Oki Electric, which has been manufacturing mainframes for 50 years. --Ancheta Wis 08:49, 1 August 2007 (UTC)

Size of the market
I remember reading/hearing that there are 30,000 remaining mainframe sites ; IBM recently revealed statistics for 10,000 members of that market. Comments, anyone? --Ancheta Wis 09:31, 1 August 2007 (UTC)

ENIAC
I removed the following sentence: "For example, ENIAC remained in continuous operation from 1947 to 1955." It makes no sense since Eniac had problem with reability due to the high number of vacuum tubes. There ar no mentioning of this in the ENIAC article. I even doubt that ENIAC could be defined ass a mainframe even if it definitely had the size so did it not have the other characteristics of a mainframe. 82.209.130.109 (talk) 20:53, 26 February 2008 (UTC)

Whu?
What on earth is "[decreasing] Linux and Java processor prices by about 25%" meant to mean? --Fuzzie (talk) 02:27, 8 March 2006 (UTC)


 * Belated answer, but: IBM mainframes can be fitted with special-purpose processing units that are specific to Java, or to Linux.  These units can't be used to execute mainframe software in general, only Java or Linux.  (The limitation is enforced mostly by what PC folks would call firmware.)  The big win is that software licenses are often priced in terms of the number of general-purpose processing units you have; so if you buy Java processors, or Linux processors, you can increase your capacity for Java or Linux without increasing the price of your other software.  The special processors are also cheaper, even though they actually include more technology (not just a CPU, but also the stuff needed to limit your use of it).  It's stupid, but that's the way the game is played.67.158.77.34 04:04, 6 November 2006 (UTC)

'IBM mainframes can be fitted with special-purpose processing units that are specific to Java, or to Linux' - read x86 processors to do work to connect IBM's mainframes to the modern world. Funnily enough theres a similar SNA over TCP processor too. —Preceding unsigned comment added by 88.110.45.91 (talk) 16:11, 27 June 2009 (UTC)
 * Incorrect. The "special" processors are actually the regular System z processors.  What makes them "special" is that IBM will disable certain functions in the chip that causes them to not to process z/OS workloads.  In exchange for this 'disabling' the processors are sold/leased at a cheaper rate.  --TreyGeek (talk) 21:03, 27 June 2009 (UTC)

Too much POV
Give a careful read to this article, especially the point of view - so many sentences need to be rewritten but I dare not for fear of my own (relative) ignorance of mainframe computers. 71.116.217.242 20:25, 30 March 2006 (UTC)


 * I agree. The writer seems to be making an effort to be balanced, but it comes off awkwardly. There are too many qualifications, needless adjectives, and speculation ("...that strategy should help Unisys and HP improve their margins..."). A couple more examples:
 * "... proven reliability, high-quality technical support, top-notch security..."
 * "When discussing the mainframe market it's no exaggeration to say that IBM defines it."


 * This needs a rewrite by somebody familiar with the area. User:24.60.149.101 (talk • contribs)


 * ... proven reliability, high-quality technical support, top-notch security...
 * That's a true statement.
 * It very well may be, but the flourish of it sounds more like marketing than reference.--Tjkroh 16:06, 23 May 2006 (UTC)


 * When discussing the mainframe market it's no exaggeration to say that IBM defines it.
 * So is that.
 * Ibid.--Tjkroh


 * And I say this having no business relationship with IBM, now or ever. But "mainframe" really is still synonymous with IBM, z/OS, etc.
 * -Whether you have a relationship with them is irrelevant to the propriety of the tone. Don't get me wrong, it was an excellent article, and I enjoyed it a great deal, but it reads like a IBM brochure. It'd be nice to have a Unisys sympathizer give it a once-over to improve the balance.--Tjkroh


 * Atlant 13:00, 28 April 2006 (UTC)


 * Who's Unisys??? ;-)


 * Atlant 17:18, 23 May 2006 (UTC)

I have just completed a heavy edit on this article, and I believe it now accurately reflects the facts about mainframes and how they fit into the rest of the computing world. Feel free to carve on it and make it better. RossPatterson 02:27, 14 November 2006 (UTC)


 * This sentence still reads very marketeering:

"To give some idea of real world experience, it is typical for a single mainframe CPU to execute the equivalent of 50, 100, or even more distributed processors' worth of business activity, depending on the workloads. Merely counting processors to compare server platforms is extremely perilous." There is a whole discussion on this subject, with the mainframe side pushing the "Mainframes do the job of billions and billions of distributed processors" and Microsoft trying to counter with numbers. I understand a neutral POV would show some specific, concrete, observable, testable situation in which a certain Mainframe CPU, with x processors running with y GHz can provide the same w benchmark as some "distributed" equivalent, such as a Sun or HP blade server with x1 processors, y1 GHz each on the same w benchmark.Ricardo Dirani 14:40, 12 September 2007 (UTC)


 * I see the section "performance" has some considerations about Mainframe CPUs not differing wildly from what is current. I see an impressive *absense* of details about Mainframe CPUs. I could find only *one* source on google quoting the clock of a mainframe CPU:

http://freespace.virgin.net/roy.longbottom/mips.htm#anchorIBMb


 * Manufacturer     No. of    OS       MHz    MIPS    MAX   Type Year  Cost
 * Processor         CPUs  CPU chip           Claim  MFLOPS            £=GBP


 * IBM                     OS Z/OS
 * Z Series 64 bit
 * 2064 Z900 1C1      1      CMOS      769     250           MF  2000  $500K


 * 2084 Z990 301      1      CMOS      1205    450           MF  2003 $1.05M


 * 2084 Z990 332     32      CMOS      1205   9060           MF  2003 $15.4M


 * Compare with an Intel and an AMD systems from around the same time:


 * Manufacturer     No. of    OS       MHz    MIPS    MAX   Type Year  Cost
 * Processor         CPUs  CPU chip           Claim  MFLOPS            £=GBP


 * Intel
 * Pentium 4          1    Pentium 4  3800    3783           Ch  2004


 * AMD

Ricardo Dirani 18:39, 12 September 2007 (UTC)
 * Athlon 64          1    Athlon 64  2600    4314           Ch  2004


 * CPU MegaFLOPs are only one measure of a computer's capabilities. It's even difficult to come up with a fair benchmark comparing mainframes to server systems, since each is better at some things than the other is, so any benchmark would have to be slanted one way or another. Mainframes should do what they do best, and servers should do what they do best. Competition between them shouldn't be an issue. SFFrog (talk) 07:38, 21 July 2009 (UTC)

Why did mainframes survive the predicted doom?
They're big and they're very expensive, how come modern day servers with all their claims to be 'autonomic' and with their improved qualities of service, haven't killed the mainframe for good?

Is it just a case of it being too expensive and risky to transfer over to the servers that companies stick with mainframes?

Are there any scenarios that a company would move over to a mainframe solution from a server solution?

Any hard evidence to show the benefits of Mainframes over servers?

How long before they actually die out? --80.40.60.154 18:09, 20 Sep 2004 (UTC)


 * While I am not an expert on this, I believe there are a few factors. One is that mainframes tend to survive in places where speed of data access is critical. Mainframes can be designed to optimise data throughput, in a way that Intel boxes cannot be. Secondly reliability. One mainframe replaces several Intel servers; this basicly means fewer possible causes of failure. Thirdly, support. Again if one mainframe replaces many Intel boxes that's less maintenence work to do. Fourthly one stop shopping. If your IBM mainframe breaks, IBM will come and fix it for you. DJ Clayworth 18:19, 20 Sep 2004 (UTC)


 * Just a quick comment on performance - every bank and credit card clearing house in the world uses some sort of mainframe for processing transactions. There is a no type of "server", "farm" or "cloud" that could handle the sheer volume of transactions on black Friday for instance. —Preceding unsigned comment added by 12.234.201.253 (talk) 22:48, 23 November 2009 (UTC)

One cannot ignore simple inertia and platform lock-in. COBOL is not as portable as most literatures makes out. Vendors have, over the years (given the extremely slow rate of change in standards... 1960, 1974, 1985, 2002) have implemented their own extensions. Companies with literally millions of lines of COBOL simply cannot take it off. Additionally, transactionality/unit of work management is important. Without an adequate and performant two-phase commit ability, slowly migrating systems away from the mainframe is difficult. One can think of the mainframe as a nail in the centre of a system, with attempts to migrate away being a rubber band fastened around the nail. Problems often mean that the path of least resistance (if not ultimately the optimal path) is still on the mainframe, being written in COBOL. 7 Jun 2005. —Preceding unsigned comment added by 82.41.212.98 (talk • contribs) 16:42, 9 June 2005

Highly unusual comment been added at the foot of the article, re: Unix having no standards, which is only partially true from the viewpoint of twenty years ago. Mainframe 'standards' are only based on the fact that they are proprietary and hence follow a 'standard' of sorts. A subtle comment indicating that there are 'no known viruses on the mainframe' it seems to me is intended to indicate some sort of design superiority over Unix/PCs/Windows etc. If it were that easy, everyone would be doing it. It seems to me there are various reasons why a virus is unlikely on a mainframe.


 * They cost large sums of money (into the millions of dollars) to buy one; hence, hackers typically don't have one in their bedroom
 * They operate on closed networks
 * There is relatively little documentation on the OS

MetalMickey 20:07, 6 August 2005 (UTC)


 * Whoa! IBM's documentation is unbelievably extensive, even though it's harder to find what you need ever since IBM went to distributing documentation on CDs in the 1990's. For decades IBM was the world's second largest publisher, exceeded only by the US Government Printing Office. SFFrog (talk) 06:53, 21 July 2009 (UTC)

A related question is "Why did the conventional wisdom come to believe that mainframes were doomed?" Wikismile 13:53, 12 August 2005 (UTC)

"Undoubtedly debates will continue about the mainframe's value ... The debate began with the 1964 introduction of the IBM System/360 and has continued for over 40 years." I don't recall any debate at all during most of this period. Even during the 1970s, minicomputers represented an expansion of computization rather than a threat to mainframes. Where is the evidence of such debates? --Wikismile 15:20, 26 October 2005 (UTC)

I worked for ICL in the late 1970's and 80's and I do remember debate taking place particularly after the first desktop microcomputers started to appear in the mid-eighties. Part of the reason why mainframes survive is the major costs involved in moving systems. If your system works don't fix it is the attitude. The costs involved in moving systems don't only include the relatively small costs of new hardware and software. They include converting data, retraining staff (particularly management) in the new systems, finding work-arounds for the stuff the old system did but the new system doesn't do. Wilmot1 03:06, 1 March 2007 (UTC)

I've seen no mention of scaling here. With very few exceptions (such as Google's amazing work), mainframes scale better than regular servers or server clusters due to their hardware architecture. Larger amounts of data can easily be stored and accessed more efficiently on a mainframe. Also, the large number of I/O channels makes data availability times shorter.

I believe, however, the main reason mainframes will never die is their reliability and security when compared to any existing alternatives. Their role has changed from what they were in the past. Maybe it's best just to think of mainframes as huge, dependable servers for many applications. SFFrog (talk) 06:53, 21 July 2009 (UTC)

Supercomputers
comparison with supercomputers: basic idea: supercomputers are for compute tasks, mainframes are for reliability and io problems. actually the difference is pretty hard-and-fast one in my opinion. if you look at products marketed by suprecomputer companies and mainframe companies, you see at least the following differences: supercomputers are geared towards doing computations instead of organising data and shuffling around io. mainframes are engineered for reliability, availability and serviceablity. mainframes typically have relatively feeble cpu power compared to their contemporary high performance computing platforms. mainframes are designed to reliable transaction processing, whereas supercomputers are designed to churn through computative workloads with i/o systems fast enough not to bog down the computation business.

there is a saying: a supercomputer is a machine that converts an i/o bound problem to a compute bound problem.

Isn't it "a supercomputer is a machine that converts an computer bound problem to an i/o bound problem."??? 109.90.35.132 (talk) 16:04, 25 February 2010 (UTC)

sorry i'm not much good with producing wikipedia-quality article text.. —Preceding unsigned comment added by 213.139.167.121 (talk • contribs) 20:17, 5 May 2004


 * Nice job, it captures the essence very well. RossPatterson 02:42, 14 November 2006 (UTC)

Redundancy
Removed this:


 * The internal redundancy of these computers can be such that, in at least one reported case, technicians could move one from one site to another by disassembling it piece by piece, and reassembling it at the new site, whilst leaving the machines running. The switchover in this example took place entirely transparently.

I know this is in the Jargon File, but on reflection a better and more specific source would be nice. --Robert Merkel 05:11, 17 Feb 2004 (UTC)
 * On top of that, it's a bunch of malarkey. It just doesn't happen, and never has. RossPatterson 02:39, 14 November 2006 (UTC)
 * Actually it has happened. A 4 node series 39 level 80 ICL VME mainframe had this done to it in the late 80's or early 90's. Two nodes were configured out, then switched off and moved from one site to another (Hitchin to Bracknell I believe but could be wrong). Once there they were set up with a new set of disks and other controllers that had been purchased specifically to extend the machine. The new machine was then powered up and configured, the services transparently moved over the x25 network, and then the rump of the original machine was switched off. These two nodes and most of the other hardware was then also moved, set up, powered up and reconfigured in. One of the joys of VME is that it can dynamically configure multiple nodes into a single machine running one set of work, or into more than one machine each running its own workload. In essence that's what happened here. The machine was reconfigured into 2, then one of those machines was switched off and moved, switched on in the new location, took over the work from the original, and then the balance of the original was switched off and moved before being brought up as a second machine and then configured back into a single machine. Starfiend (talk) 19:23, 2 March 2010 (UTC)

Linux references
I'm not trying to arbitrarily nuke linux references from the article, it's not like it's not news that IBM supports it on their big iron. I deleted the references because that entire section looked like it was written just to put in OMG IT RUNS LINUX!!! Even the title makes no sense whatsoever. And the fact that IBM promotes Linux on the mainframe is not reason enough to add all that, on the contrary, it looks even more like a press release from them. WP is not a vehicle for commercials about technology.

I suggest adding a separate section, correctly attributed and sourced, about Linux on IBM mainframes (and only in IBM ones, since no other manufacturers support it). And something a bit more neutral than "IBM sells lots of mainframes. They run Linux. Some even run z/OS" would be indicated. §FreeRangeFrog 07:54, 18 October 2008 (UTC)
 * IMHO, this article is about mainframe computers in general, and not IBM mainframes. If we look at the articles on the IBM mainframes we'll see information about Linux on System z, and that is appropriate.  I will agree with FreeRangeFrog and their edits.  Linux in terms of mainframes in general deserves a passing reference. --TreyGeek (talk) 15:03, 18 October 2008 (UTC)
 * Also, is it really true that only IBM's mainframes support linux? I'm not actually sure that it is. I'm fairly certain the ICL/Fujitsu Supernova mainframe range will also run linux. —Preceding unsigned comment added by Starfiend (talk • contribs) 23:03, 5 March 2010 (UTC)

Amdahl
I'm not really sure why I started reading this article, but I was surprised that there's no mention of Amdahl. Modal Jig (talk) 19:31, 15 March 2010 (UTC)

Mainframe language breakdown
Where does the breakdown of programming languages used come from? I know that a lot of mainframe programming was done in Assembler, and much of that code is still out there being maintained. After Assembler, COBOL was popular, and probably is the most popular high-level language out in mainframe-land; but is it 90% of development? I doubt it. Nowadays, there is a lot of development in Java (especially on Websphere [which as far as I can tell is an IBM mainframe version of Apache], but also in CICS). C/C++ has been used for years (I know because I've done some). Then there are all the scripting languages: REXX, EXEC, CLIST, even JCL. —Preceding unsigned comment added by 66.150.41.150 (talk • contribs) 18:28, 12 July 2004


 * Got it from a bloke at work the other day who's been working on mainframes for 35 years, other than that i can't really justify the figure given. Even though a lot of the new code is in Java, i believe thats only because its enjoying its 15 minutes of fame.  I guess the reason COBOL perseveres so much would be down to the 20 year nearly competition free spate it enjoyed as the main language on the mainframe.  To this day a lot of changes and expansion code on mainframes would be done in COBOL while they have the employees with the skills.  The complexity of COBOL is not far off assembly, and i believe a direct descendant of tape code
 * If you think we should drop that statement i have no problem, i will try and get some firmer evidence later this week.Pluke 23:53, 12 Jul 2004 (UTC)


 * Have adjusted the figures accordingly from talking to some serious mainframe evangelists at work. Trying to find the figure for lines of new COBOL code written for mainframes each year; it's supposed to be huge! Pluke 20:41, 14 Jul 2004 (UTC)

I'd tend to agree with Pluke on this. I've spent 30+ years as an IBM mainframe Systems Programmer, working in many companies in the San Francisco Bay Area (I'm an ex-IBMer and independent consultant), so I think my experiences count. Although there have always been plenty of languages supported under OS/360 and all its successors including today's z/OS, COBOL far and away has been the most-used language for applications programming. Originally, the operating System was written entirely in Assembler for efficiency (Assembler is one-for-one with machine code, whereas COBOL is a high-level language in which each instruction coded generates multiple line of Assembler/machine code). I can't agree that COBOL is even close to Assembler in complexity -- I know both well.

In the 1970's IBM began writing some of the SVS (and later, MVS) code in PL/S (Programming Language/Systems), a language that had a syntax similar to PL/I's. It generated Assembler Code and left the PL/S code in the listing as comments. It was only used within IBM.


 * PL/S dates from the 1960's, when it was called BSL. Large parts of TSO in OS/360 were written in BSL.


 * Also, PL/S and assembler were not mutually exclusive, because PL/S included the ability to embed assembler code in the source and copy it into the output. Shmuel (Seymour J.) Metz Username:Chatul (talk) 16:34, 30 August 2010 (UTC)

Assembler was seldom used as an applications language. The only times I ever saw it were in a few modules that were used to do complex functions that were executed extensively and/or were time-dependent in their execution. Systems programmers always wrote this code, since applications programmers almost never learned Assembler.

No place I've ever worked uses C or C++ on the mainframe, although there was a C compiler generally available starting in the late 1990's on MVS/ESA or OS/390. The only time I ever write in either is for a PC. Since many people have learned C or C++ in college computer science courses (particularly since it's THE language in UNIX and its offshoots), there is undoubtedly more use of those languages on mainframes now than in the past, just not in my experience in large corporations.

EXEC is used only in VM, CLISTs are only used in TSO under MVS and its successors, REXX was first used in VM, then became available for TSO as well. Calling JCL (Job Control Language) a scripting language is really stretching the definition of scripting, but I can understand the thinking. SFFrog (talk) 06:38, 21 July 2009 (UTC)

Other Manufacturers
ICL should be included in the list of mainframe manufacturers and their 2900 series in the list of ranges. Wilmot1 02:55, 1 March 2007 (UTC)


 * Just about any manufacturer of magnetic tape and vacuum tubes had a role in the computers of the 1940s.  Shouldn't there be an article about RCA and Westinghouse and Columbia Broadcasting System, not to mention Bell Telephone Company?    As for flexibility in automating construction lines for aircrafts, tanks and jeeps, this being a matter of efficiency, times saving, and profit, the award of contracts to large companies by the US Armed Forces needs to be described.  Dexter Nextnumber (talk) 21:42, 14 December 2009 (UTC)


 * Manufacturing components isn't relevant; you wouldn't list a ball-bearing company as an auto manufacturer. I'm nor aware of any role played by CBS.


 * The companies that I believe should be discussed are
 * Bendix
 * B.U.L.L.
 * Burroughs
 * CDC
 * Elliot
 * Ferranti
 * GE
 * Honeywell
 * NCR
 * Philco
 * RCA
 * SDS
 * Sylvania
 * Unisys as the successor to Burroughs and UNIVAC
 * UNIVAC Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:09, 30 August 2010 (UTC)

PDPs
Some PDP's were mainframes.

Year 18-bitters     12-bitters             16-bitters                36-bitters

1960   PDP-1 1961     |                                                               \ 1962    PDP-4 <--- LINC                                            \ 1963     |        PDP-5   \    \                                           | 1964   PDP-7        |      \    \                                       PDP-6 1965     |        PDP-8 --\ |    \                                         | 1966     |        PDP-8/S LINC-8  |                                        | 1967     |          |       |     |                                     PDP-10 KA10 1968   PDP-9      PDP-8/I,L |     |                                        | 1969     |          |     PDP-12  |                                        | 1970   PDP-15       |           PDP-14    PDP-11(/20)                      | 1971     |        PDP-8/E                  /   |  \                        | 1972   PDP-15/76  PDP-8/M           PDP-11/05  |  PDP-11/45 --          PDP-10 KI10 1973                |             /   |       PDP-11/40  |    \            | 1974                |            /    |            |     |     \           | 1975               PDP-8/A   PDP-11/03 PDP-11/04    |     |  PDP-11/70   PDP-10 KL10 1976                |                 PDP-11/34    | PDP-11/55  |          | 1977              VT78                   |    PDP-11/60         |          | 1978                                  PDP-11/34C            VAX-11/780  PDP-10 KS10

In a nutshell all the 36-bitters were mainframes and the rest weren't.


 * I'm not sure that I'd classify the PDP-6 as a mainframe. The larger PDP-10's certainly were, and I don't understand why you don't count the VAX as a mainframe. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:09, 30 August 2010 (UTC)

DEC pulled out of the mainframe business before completing the PDP-10 KC10 and jilted their whole mainframe customer base (including CompuServe and MCI/Tymenet) which marks the beginning of the end for DEC since many of those customers moved to UNIX not VMS where DEC couldn't pull the rug out from under them again. —Preceding unsigned comment added by 66.92.164.43 (talk • contribs) 13:06, 21 December 2002

The PDP-10 is described as a mainframe on its page. The IBM-heavy tone of the article is misleading. Trashbird1240 (talk) 17:17, 13 February 2008 (UTC)


 * Having been a devoted user of the KL-10 (Tenex), the 2060 (TOPS-20), and occasionally some of the ITS-modified KL-10s, as well as most IBM 3x0 systems from the 370/158 through the early zSeries (MVT, MVS, and VM/CMS), I have to disagree. Kubanczyk's description of the PDP-10 as a mainframe is just plain wrong.  They didn't have the well-balanced complement of I/O, memory, and CPU that characterize a mainframe, and they were never considered to be high-end processors (another mainframe characteristic until Seymour Cray's supercomputers dominated that market).  RossPatterson (talk) 03:16, 14 February 2008 (UTC)


 * Having been called to the board, I'm adding a source to the PDP-10 article. This is of course per WP:RS. --Kubanczyk (talk) 10:23, 14 February 2008 (UTC)


 * DEC's marketing literature isn't much of a reference, but I won't argue the point. RossPatterson (talk) 00:34, 15 February 2008 (UTC)

Performance
Finally, I get to ask an encyclopedia a question. What sort of performance do mainframe computers have? CPU speed (GHz)? Number of processors? Number of MIPS? Quantity of memory? I know that there will be a range of values, but it would be good to see either a range of values or particular values for one mainframe type. Thanks. Mjm1964 16:41, 22 Jun 2005 (UTC)


 * Generally speaking, "mainframe" computers contain about the same speeds of processors as you'll see in other computers; everything is built out of the same ULSI semiconductor processes so the speeds and scales of the processors naturally fall out of that. One difference you will see is that certain processors (such as those used in the IBM z/OS computers) are actually internally-duplicated processors running in lockstep; if the two processors disagree on the results of a machine-language instruction, they fault it out to the service processor to figure out which one got it right,disable the one that got it wrong, call the Field Service Engineer, and then continue as if nothing had happened using only the non-faulted half of the chip. Such processors probably clock a little slower than your average bear, but they essentially never fail completely so the trade-off in speed is thought to be well worth it.


 * With regard to memory, mainframes tend to contain a lot of it but not many binary orders of magnitude more than is current at any given time. I'd guess that it tops out at several hundreds of GB right now. With regard to processors, it depends on the architecture and just what you're calling a mainframe. I *THINK* z/OS systems are still pretty low in processor count (8? 16?) while PowerPC-, Alpha-, and SPARC-based systems range up in the 128-1024 processor range.


 * If we were to include this sort of data in the article, I think we'd have to consider very carefully what represents a "mainframe" versus what represents a "supercomputer" versus what just represents a state-of-the-art big Unix server. If we're going to go down that route, I'd suggest that we use the IBM z/OS series as the reference point (and it's certainly well-enough described often-enough in editions of the IBM Journal of Research and Development that we could draw good, reliable data from there).


 * Atlant 18:28, 22 Jun 2005 (UTC)


 * I agree. z/OS is clearly the reference point. I think IBM are the only company making machines that run z/OS nowadays. IBM's big Unix machines do in many ways qualify as 'mainframes' in the sense that they've got a huge I/O capacity; however, I think the best definition of 'mainframe' is a machine that runs OS/360->z/OS. There's not really a scientific definition; the term is more an accident of history, rather than a well thought through definition. Incidentally, mainframes run a max of 256GB presently. It's also worth pointing out that these machines have been largely resilient to changes in thinking around what a good instruction set resembles, since they have to still run programs written for the first S/360 in 1964. MetalMickey 19:45, 23 Jun 2005 (UTC)


 * There have been several additions to the original S/360 instruction set in the evolution to the z Series mainframes. The S/360 instruction set is a huge subset of the current one because it was inclusive of almost anything that could ever be done with the base S/360 architecture other than special instructions to enable easier use of virtual memory.  SFFrog (talk) 07:26, 21 July 2009 (UTC)


 * An important aspect of performance is the I/O bandwidth. A current IBM processor can have a huge number of fiber channels. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:09, 30 August 2010 (UTC)

Statistics
I'm a little confused about the paragraph in the statistics section which says that people think mainframes are expensive, but the reality is different. First, $50,000 still seems expensive. Secondly, I'm not sure where the judgement that it isn't expensive came from; the only thing the reference cited seems to backup is that the OS costs about $1,500. Rnb (talk) 15:25, 2 July 2009 (UTC)
 * While $50k (or $100k for the baseline IBM System z10 Business Class mainframe) is expensive for individuals, it is not expensive for the businesses and corporations who use them. One thing that is not explained (and would be difficult to explain in a short amount of space) is that while the initial outlay for the hardware and software may be higher than an x86 based server [farm], expenses will be cheaper over the life of the system (particularly when factoring in costs to power the machine, maintain & support it, provide adequate cooling, etc).  A quick example is Nationwide Insurance moved several hundred of their Linux x86 servers onto virtualized Linux images on a System z9.  They are saving in the area of $30M per year with the mainframe vs the x86 server farm.
 * As for the statistics section as a whole, it should probably be removed (or its presentation heavily edited) in favor of WP:TRIVIA. --TreyGeek (talk) 18:07, 2 July 2009 (UTC)
 * Got it. Thanks for the info.  I guess my only real input would be to either, as you say, have the section removed, or have it re-written so that it sounds less like someone's opinion and more like it's a comparative fact.  Thanks again. Rnb (talk) 18:10, 2 July 2009 (UTC)


 * I don't know where you can buy a mainframe for $50,000 - back in the late 1980s/early 1990s I worked for a UK Facilities Management company at a site that had five IBM 3090's at IIRC, around £3,000,000 each. We also had two IBM AS/400's, three DEC Vax's and two Honeywell-Bull DPS-7's and a couple of Honeywell-Bull DPS-6 Model 40's. I won't say who the clients were but you've almost certainly heard of them. Back then we had also a few PCs but they were only used for writing internal reports and we all looked down on them as being a bit 'Mickey Mouse'. IIRC, they were all running DOS so it was all Command Line stuff, but then again so were the mainframes.


 * The main reasons for buying a mainframe back then was that apart from the lack of power of the then-average PC, the mainframe and mini manufacturers provided excellent (for a price) backup and support. And the Operating Systems and programs just worked almost all the time. And most mainframe and mini buyers would have dedicated in-house programmers available to fix anything when it didn't. So it wasn't just the initial cost that was expensive, it was the overall support structure that needed to go with it that cost the most. Generally though, if you wanted the best and most reliable data processing service, i.e., your business was heavily reliant on data processing and depended on it - such as banking and insurance, then that was the way to go. It probably still is. —Preceding unsigned comment added by 86.112.41.48 (talk) 21:30, 24 October 2010 (UTC)

POV
This seems like so much marketing material from IBM rather than a reflection of the current status of mainframes. The article needs to be heavily edited for balanced point-of-view, substantiated claims, references with a formal, neutral tone. —Preceding unsigned comment added by 67.180.83.26 (talk • contribs) 23:22, 19 October 2006
 * Fixed, see below. RossPatterson 02:43, 14 November 2006 (UTC)
 * I don't think it's fixed. Amnd, for start, you were one of the main editors of the article, and are part of the oldest IBM usergroup, with links to the enterprise... how could you have fixed it? You can't be part of the solution if you are a part of the problem. --190.174.81.172 (talk) 04:46, 29 October 2009 (UTC)


 * I don't agree that an editor is automatically unable to write objectively about material he has worked with, and, in fact, I don't see how you will get accuracy without such an editor. However, I agree that the article needs a lot more material on other vendors, and not just the seven dwarfs either. Shmuel (Seymour J.) Metz Username:Chatul (talk) 16:19, 30 August 2010 (UTC)


 * Dang sure. This is all POV like an IBM ad. —Preceding unsigned comment added by 65.114.207.226 (talk) 19:36, 6 December 2009 (UTC)

Agreed, this reads like an advertisement for IBM. (Drn8 (talk) 17:24, 22 August 2011 (UTC))

Dumb terminals
mainframes used non-"dumb" terminals, with some editing and form functionality in the terminal itself. in "dumb terminals" each keypress was transmitted to the host, which updated the display accordingly. the "dumb terminal" was coined in contrast to the terminals used before with mainframes that had more smarts. —Preceding unsigned comment added by 213.139.167.121 (talk • contribs) 20:01, 5 May 2004


 * Fixed. You're absolutely right - the 3270 family of terminals is anything but "dumb". RossPatterson 02:40, 14 November 2006 (UTC)

No he/she isn't "absolutely right". Just because "each keypress" is not "transmitted to the host" (called echoplex) doesn't mean that a terminal is not dumb. It is simply that 3270 terminals were block oriented rather than character oriented; the original 3270 terminals such as 3277 and later the 3278s had no "intelligence in the head" and so were dumb. Not all 3270 terminals are dumb however: a 3277 is dumb (ANR protocol) as is a 3278 (CUT protocol) but a 3270PC/GX is far from dumb (DFT Protocol). Terry. Ex-IBM 3270 specialist. — Preceding unsigned comment added by 86.134.45.70 (talk) 17:24, 24 August 2011 (UTC)

Deleted POV and weasel word templates
I deleted the very old POV and weasel word templates. I agree that the article could be improved by editing it to a more neutral point of view and use of more concrete lanugage, and also by adding (and following) WP: reliable sources. However, the templates were 2 and 2½ years old, respectively. That is more than ample time for editing. Feel free to improve the article by editing it. However, hanging templates and complaining on the Talk page do not improve the article.—Finell 19:43, 14 January 2012 (UTC)

Characteristics of early mainframes
The description of mainframes in the 1960's being mostly batch is seriously overstated. While batch was common, remote terminals were used interactively even in the 1950's, and the 1960's was the period when online applications emerged into the commercial marketplace.

Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:39, 8 July 2012 (UTC)

Speculation on evolution of the word mainframe
This is pure guesswork, but the use of the word to mean "big honkin' MIS computer" has always puzzled me. Here's my speculation. Comments welcome. I always used it to mean "the bay or bays containing the CPU in any floorstanding computer," and was annoyed one day when someone said "that's not what it means; it means a computer that costs more than a million dollars." Here's what I'm guessing took place.


 * First, mainframe means "the bay or bays containing the CPU in any floorstanding computer."
 * Third-party vendors of plug-compatible peripherals for IBM 360-series computers don't want to mention IBM by name, so start referring to IBM with the circumlocution "the mainframe vendor." Or, people need a neutral or indirect way to say "IBM and/or Amdahl." So the phrase "mainframe vendor" comes into use.
 * Extrapolating from this use, "a mainframe vendor" is soon taken to mean "manufacturer of big honkin' MIS computers." Or "IBM and the seven dwarfs."
 * One more cognitive step is needed, though, because IBM of course made computers of all sizes. So the next step is "mainframe" comes to mean "the kind of computer most commonly associated with IBM" = "big honkin' MIS computer."

Thoughts? [[User:Dpbsmith|Dpbsmith (talk)]]


 * The jargon file says the term refers to the systems' enourmous outer casing. --Apantomimehorse 08:34, 16 August 2006 (UTC)


 * That's exactly what he means. The "enourmous outer casing" is the "bay(s) containing the CPU". For example, the System/360 model 168-III's CPU covered a 250 square foot area, and each pair of I/O channels held down another 25 square feet of floor tiles. We'll probably never know, though.  The jargon file is a lousy source in this case - it came from the opposing camp.  RossPatterson 02:50, 14 November 2006 (UTC)


 * Really big, honkin' supercomputers existed in science fiction in 1951, as can be seen in the sf story written by J. T. McIntosh but did the word 'mainframe' really exist in the 1950s?   How do you know whether the word existed any earlier, perhaps in the 1940s?   Somebody needs to locate a source so we can support the theory with a reference.   And speaking of which, if an external site doesn't load up in 5 or 6 minutes, there's probably something wrong with it.   (For which reason I decided it best to delete the footnote to somebody's theory that IBM had anything to do with the word "mainframe.")   The word "mainframe" looks like it was borrowed from the carpentry trade or mining trade.   If you have ever had to frame a house or a barn, the construction trade looks also like a highly appealing point of origin.   On the other hand, if you have drawn doggies by tugging or dogging such so-called carts up and down a mine shaft, it is possible that the mining trade might have had something to do with it, if only because doggies were about 4 feet tall, 8 feet long, and 4 feet wide in size.   Finally, very large radios were often (always) made with vacuum tubes inside them, and the early computers of the 1940s and 1950s all had lots of vacuum tubes in them.   The word "mainframe" in that sense could have been related to the size of radio cabinets of the time.  And if that is the case, cabinetmakers might have had a lot to do with the rise of the term.  —Preceding unsigned comment added by Dexter Nextnumber (talk • contribs) 23:25, 13 December 2009 (UTC)


 * THIS (pictured right) is the main frame of Colossus, an early computer. The name 'main frame' refers to the section of the computer where the bulk of the electronics were located, as opposed to the peripherals, such as paper tape readers, etc. As you can see, the computer is without any form of case, so it is open to the air for cooling and access, as the thermionic valves (vacuum tubes) are great generators of heat, as well as needing occasional replacement. So the machine is constructed around a frame.

—Preceding unsigned comment added by 86.112.41.48 (talk) 20:46, 24 October 2010 (UTC)

I found a website that has some interesting discussion on the etymology of the word mainframe. Not a good RS citation but maybe a good place to add perspective to a search for origins. 159.83.196.19 (talk) 00:11, 16 January 2013 (UTC)

MSU
An MSU is a unit of CPU, I/O and storage resource use, not a rate. Further, IBM adjusts the nominal MSU capacities of various models for marketing purposes, so the number of MSUs consumed by the same job will differ depending on which model it runs on and thus MSU's/s does not allow comparing speeds between processors. Shmuel (Seymour J.) Metz Username:Chatul (talk) 19:13, 2 July 2013 (UTC)

What is it exactly?
This article meanders on through endless "some say, while others say" and "a mainframe does this, sometimes, but not always, and are different from servers, although not necessarily". After reading through it I don't have any idea what a mainframe actually is due to all the weasel words and hedging. - Keith D. Tyler &para; 21:30, 2 January 2010 (UTC)
 * That's because the article contains so much falsehoods and half truths that it should be deleted and started all over again, and not by the IBM sales team. Mainframe is not, and never has been, an IBM only term. Starfiend (talk) 22:57, 5 March 2010 (UTC)
 * I was wondering what is a real defining feature of the mainframe computer, instead of repeating "old", "big" and "blue" adjectives. To me it seems that particular reliance on offloading I/O operations to devices by the means of Channel I/O, where a relatively high-level standard interface is used for different I/O devices; often combined with record-level processing. In those terms, modern PC architecture with Bus mastering and offloading I/O operations to intelligent devices, many of which run its own firmware (for example in Universal Serial Bus) is getting more and more mainframe-like in a way; except that it is not standardized (i.e. there is no common channel for all I/O) and there are no channel programs. I presume this has been somehow addressed in the literature already, somehow? « Saper // @talk »  17:34, 9 July 2013 (UTC)


 * I would say that the definitions of various classes of computers are like the Supreme Court ruling on pornography: I may not be able to define it but I know it when I see it. I'd say that the important aspects for contemporary mainframes are:


 * Reliability, Availability and Serviceability (RAS), e.g., ECC memory, software recovery from hardware and software failures.
 * Ability to support large numbers of peripher devices with high bandwidth.
 * Ability to support large memories and large numbers of processors.


 * To some extent the term is a moving target; a machine that was considered a mainframe in 1958 would have been considered junk in 1968.


 * The article badly needs material on other vendors. Shmuel (Seymour J.) Metz Username:Chatul (talk) 21:13, 10 July 2013 (UTC)

This article was written by IBM's marketing department
I would strongly advise that this article be re-written. It is clearly a product of IBM's marketing efforts. There are numerous self serving statements about IBM being the leader in mainframe technology and an attempt to marginalise the competition. —Preceding unsigned comment added by 70.27.199.213 (talk • contribs) 01:45, 3 October 2006


 * Yeah, it sure reads like it. Fixed. RossPatterson 03:04, 14 November 2006 (UTC)
 * Still think Unisys need to sort their act out to steer some more truth in. IBM wasn't the first, and certainly not the best. More Univac/Sperry to show the lead and feature comparison of Burroughs/MCP vs IBM's AIX shows just how advanced the 'underdog' was. Also, do the Sun Fire 10k and 25k ystems count? —The preceding unsigned comment was added by 80.47.252.243 (talk) 00:54, 14 February 2007 (UTC).

I was big Univac 1100 mainframe fan -- still am in some ways. As employees, we could not believe how often our product was denigrated because its operating system wasn't exactly like Unix or exactly like what ran IBM's big iron. I suppose there is a human factor that rejects whatever one is unfamiliar with. Still, Univac's 1100 operating system was built to run and did run an efficient mixture of real-time, timesharing, transaction, and batch jobs. Some IBM mainframes required nonsensical software swapouts in the event the customer decided to upgrade his rotating magnetic media. Some IBM software had to run dedicated to a single, physical CPU. (Symmetric multiprocessing? What's that?)  The IBM shell was the sole software component with the ability to create and attach a file to a job. So, to compile a Fortran "deck," the shell was required to be instructed to create spill files that were of internal interest exclusively to the Fortran compiler. This could be hidden, somewhat, from the end user. But it smacks of a half-baked idea when what you want is to give the Fortran compiler the ability to create its own spill file as and when it is required. Unix, because of its free distribution among academics, had the patina of being the world's finest operating system, because only the best graduate students ever had an opportunity to enhance it. In comparison to any other operating system, especially commercial ones, Unix was "the best" because of its supposed superiority of design. One design feature of Unix often touted is the ability or requirement to write applications as chains of small, buggy applications linked together via shell script punctuation. Another terrific(?) idea was to make every peripheral device in some way analogous to a teletypewriter. Hence, disk files contain records that are terminated with an ASCII carriage return, or linefeed, or both, which must be catered to by programs reading files that do not subscribe to the concept of communications control characters embedded in data. One man's meat is another man's poison. But Univac managed to avoid most of the nonsense that others cavalierly put into their design and who continue to live with to the present day.--72.75.86.145 (talk) 02:30, 27 December 2007 (UTC)

The article is still written as an IBM advertisement. Just look at the "Market" and "History" contents. It is shameless publicity, dull and plain. And how can you define mainframe as ONLY the computers compatible with the IBM 360 line? That's almost the same as saying the only OS that counts is Microsoft Windows... --190.174.99.233 (talk) 03:53, 14 September 2009 (UTC)

When reading this article, I also had the definite impression that it was all about IBM. alex (talk) 09:24, 28 July 2014 (UTC)

Reference 7
Reference 7 is a dead link. 161.31.231.168 (talk) 00:39, 5 September 2014 (UTC)

Proposed merge with Death of the mainframe
Collection of quotes regarding mainframes. Could also be transwikied to Wikiquote. Q VVERTYVS (hm?) 17:25, 25 December 2013 (UTC)
 * It's a thought. I wanted to get the article title in since it's been such a topic of controversy in itself, but I suppose it could be a redirect to here.  On the other hand I would like to find Alsop's original editorial and quote that and more pro-and-con on the controversy. Peter Flass (talk) 21:42, 25 December 2013 (UTC)
 * It can be a subsection within the History section, with a specific redirect. That makes it easier for readers to put the discussion in context. Q VVERTYVS (hm?) 22:04, 25 December 2013 (UTC)
 * Merged. Q VVERTYVS (hm?) 10:58, 13 November 2014 (UTC)

Origin of the word mainframe
I have searched the web for thirty minutes without finding the origin of the word mainframe. Am I the only person out here who knows where the word came from?

In an old electromagnetic telephone central office, each telephone number was associated with a given physical switch. Switches were all wired back to terminal blocks on a big rack. The incoming cables also terminated in terminal blocks on the same rack. Assigning a telephone number to a specific location, i.e., to the copper pair going to that location, was a matter of making a cross connect on this rack. The rack was called the "mainframe".

I don't know exactly how the word got transferred to large computers. I would suppose it came about because at least some of this computer equipment was mounted on standard racks, as was the equipment in a telephone office.

In 1968 in the signal school at Fort Monmouth, one day when we were working on rack-mounted computer components I referred to the rack as a "mainframe". Who out there can claim a computer-related use of the word earlier than 1968?

Joseph Mansfield, 14Aug2005, Jm546

See also 19-inch rack. Jm546 17:12:48, 2005-09-06 (UTC)


 * Odd, I've always been told (by 1940s to 1960s Bell Labs and Western Electric guys) that those things were called "crossbar switches", not "mainframes".RossPatterson 02:54, 14 November 2006 (UTC)


 * All of the above. See main distribution frame, telephone exchange, crossbar switch and panel switch Jim.henderson 01:25, 5 March 2007 (UTC)


 * All of this is interesting.  So it relates to cross-connections linking up to some kind of fancy circuitbreaker?  Have you got any reliable dates to go with these terms?    I have heard of shortwave radios and radio consoles with switches to primaries and secondaries, and most of these had a "main switch" or a "main breaker" on their cases - probably a term from the electricians' trade then, rather than a telephone carrier.   Dexter Nextnumber (talk) 23:38, 13 December 2009 (UTC)

It just relates to the primary support structure of the computer, the main framework. The Oxford English Dictionary has a quote from a 1964 Honeywell Data Processing Glossary for main frame referring to the central processor and memory, or that part of the computer exclusive of input output and peripherals. They also have references to "main frame" as early as 1887 in reference to a lathe.--agr (talk) 00:27, 15 December 2009 (UTC)

My recollection of the term "mainframe" relates to the legal battles fought by IBM and third-party peripheral makers. (This would have been late 60's to early 70's.) IBM sought via patents and licensing to prevent any non-IBM hardware from being connected to an IBM system. Ultimately, IBM lost the battle, and the central processing unit, memory, control systems, etc. became the "mainframe," which was IBM's exclusive domain. System users were subsequently free to use third-party peripherals to connect to the mainframe, which was and still is considered the true core of the system. The terminology stuck, and the meaning of "mainframe" morphed to refer to large-scale systems in general. Esjones (talk) 16:20, 27 July 2013 (UTC)


 * No, in fact there have been third party channels and memory for S/360 and S/370 users. The disputes with the OEMs were over things like the repackaging in the 2319. Shmuel (Seymour J.) Metz Username:Chatul (talk) 23:09, 30 July 2013 (UTC)


 * The Colossus mentioned earlier on this page was design and constructed by GPO telephone engineers at Dollis Hill. — Preceding unsigned comment added by 95.150.100.255 (talk) 13:49, 24 April 2016 (UTC)

Recent edit
Interactive facilities were entering the mainstream in the early 1960's. They started becoming ubiquitous in the mid 1960's.

I/O channels were introduced in the 1950's and were ubiquitous in the early 1960's. It was only entry level machines that did not have them. Channels typically were not computers but comparable to DMA. In the IBM System/360, the smaller models used cycle stealing on the CPU and the larger models used hardwired outboard channels. The peripheral processors on the CDC 6600 were not I/O channels, and every PP had access to every channel. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:25, 5 July 2013 (UTC)


 * It's somewhat misleading and/or flat out false to say that interactive facilities were becoming ubiquitous in the mid 60s, depending on how you define interactive facilities and ubiquitous. Green screens and special (non-interactive) displays certainly were widely used from the mid 60s on, but tab was still predominant well into the 70s and was still widely used in the early 80s. You do say "started becoming" which is correct, but that becoming took 20 years. 76.180.168.166 (talk) 19:39, 17 August 2013 (UTC)


 * It's more than somewhat misleading and/or flat out false to equate interactive facilities with display terminals. The systems that used Selectric® typewriter and Teletype® devices were just as interactive as those using 2260 or 3270 displays.Shmuel (Seymour J.) Metz Username:Chatul (talk) 16:40, 19 August 2013 (UTC)


 * On the other hand, I wouldn't say "ubiquitous". I graduated  in  1968 from a research university, and we were batch only. Peter Flass (talk) 00:41, 20 August 2013 (UTC)
 * The sentence in question is simply wrong. Interactive computing on mainframes was cutting edge research in the early 1960s. Compatible Time-Sharing System (CTSS) was first demonstrated in 1961. BASIC began in 1964. And remember those systems were sharing a resource less powerful than an Arduino Due. Most serious mainframe work was still batch. Interactive computing did not become ubiquitous until well into the 1970s. My first job that did not involve entering programs on punch cards was in 1975. and that was for a minicomputer based system, not a mainframe.--agr (talk) 16:02, 20 August 2013 (UTC)
 * Yes, meant to say that, that initially mini's like dec and what not were leaders in non-tab input. 76.180.168.166 (talk) 19:23, 20 August 2013 (UTC)


 * 1961 is not mid 1960's, nor is early 1960s. Interactive systems were available from, e.g., CDC, DEC, GE, IBM, RCA, UNIVAC. Shmuel (Seymour J.) Metz Username:Chatul (talk) 22:05, 20 August 2013 (UTC)


 * Seems like, as for many questions, it depends on how you count. Many systems were available with interactive options, but it might be that only a small fraction shipped that way.  Do you weight based on the number shipped, or not? Gah4 (talk) 21:14, 28 June 2016 (UTC)

IBM had CRBE pretty early, which is interactive, but not time sharing. A predecessor to WYLBUR, CRBE allows one to interactively edit and submit batch jobs. Don't equate interactive and time sharing! Gah4 (talk) 21:14, 28 June 2016 (UTC)

What happened to midicomputers?
The article divides computers into "minicomputers" and "mainframes", but the industry in the 1960's and 1970's recognized three categories: minicomputers, midicomputers and mainframes. The criteria were somewhat vague, but revolved around memory capacity, price, weight and word size. In particular, the DEC PDP-7 with its 18-bit word, on which UNIX was first developed, was sometimes considered to be a midicomputer. Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:00, 3 August 2011 (UTC)
 * Are you sure about that term? I cannot find any reliable source using the term "midicomputer" in relation to minicomputers and/or mainframes.  --TreyGeek (talk) 19:06, 3 August 2011 (UTC)
 * Google Books finds an article on page 6 of "Computerworld", June 25, 1975 "Midis challenge medium-sized systems" - apparently a midi had upwards of 1 megabyte of memory and more than 500 megabytes of disk storage (put that iPad down, Junior!), and was bigger than a single-purpose "mini" but not quite as big as a general purpose system. Examples given in the article are the DEC PDP 15 (18-bit machine) and PDP 11/45, and the Hewlett Packard HP 3000 (back when Hewlett Packard thought windows were to let light in). The Google snippets referring to midicomputers seem to peter out by 1982 or so. --Wtshymanski (talk) 19:22, 3 August 2011 (UTC)


 * I don't know about midicomputers, either, but machines like VAX and the IBM 4331 might have sold as superminicomputers. That might have been partly to keep them from competing with mainframes, when they could do the same job. Gah4 (talk) 21:17, 28 June 2016 (UTC)

90% number
Is there an in-depth source for the claim that IBM has over 90% mainframe market share? All we have right now is an anti-IBM site quoting someone claiming that anti-trust action needs to be taken against IBM. 90% seems very high to me, given that Bull, Hitachi, Fujitsu, NEC, and Unisys all have strong mainframe businesses. Perhaps it refers to 90% of the US mainframe market, or 90% of /390-compatible mainframe sales? Kiralexis (talk) 00:50, 5 July 2011 (UTC)
 * That's a good question. I was able to find several sites that supported the 90% marketshare claim:  (though I'm not wild about the last two as sources).  None of them go into details about how that number is determined or whether it is US or worldwide; heck they could be pulling the number from this article.  --TreyGeek (talk) 01:16, 5 July 2011 (UTC)
 * The fact that none of those sources cites where the data is coming from is concerning to me. It's almost like the 90% figure got started somewhere, and has since just been passed around as conventional wisdom without any basis. Kiralexis (talk) 21:07, 7 July 2011 (UTC)
 * Hello.. so, 5 years later and that same poorly supported numbers is still in the article and the reference next to it is a dead link and it doesn't appear to be in the Wayback machine. I'm going to remove that number. SloppyG (talk) 18:20, 28 June 2016 (UTC)

There is no good way to use numbers like that. Is it 90% of mainframe systems shipped? Or 90% of dollars spent on mainframes? And who says which ones are mainframes? I agree, better to just remove it. Gah4 (talk) 21:19, 28 June 2016 (UTC)

MIPS
MIPS is a VAX (minicomputer) term. I have seen it applied to the VAX 11/780, which had several racks of circuit boards, several disks, and attendant terminals and drives. One MIPS meant one 11/780's standard memory, disk capacity and I/O. These are all mini-computer terms, and IBM mainframes had a different terminology and mindset. It might be useful to have a dictionary of the mainframe terms in the encyclopedia. What better place than this article? --Ancheta Wis 09:00, 2 August 2007 (UTC)
 * Actually, Million instructions per second (MIPS) was an IBM mainframe term long before there was a VAX. RossPatterson 22:30, 2 August 2007 (UTC)
 * To be clear - the first VAX, the 11/780, was introduced in 1977. That's 5 years after the IBM 370/158 and the 370/168.  IBM never quoted MIPS numbers for the 370s, but Comdisco distributed their comparison cards widely, and quoted the 158 as 1 MIPS.  Similarly, the 168 was rated at 3 MIPS.  RossPatterson (talk) 23:54, 15 February 2008 (UTC)
 * Ancheta, the dictionary is an excellent idea! --Kubanczyk (talk) 23:33, 15 February 2008 (UTC)
 * Yeah, it is! RossPatterson (talk) 23:54, 15 February 2008 (UTC)

I remember reading the 68040 CPU as used in the Mac Quadra was capable of 22 MIPS. This article says the IBM System z9 runs at 26 MIPS. Are the two really equivalent in performance? —Preceding unsigned comment added by 69.125.110.223 (talk) 19:13, 19 December 2007 (UTC)
 * Not at all, to perform the same operation it's very likely that you need several instructions on a simple CPU, whereas on a complex CP unit, the same operation, may be performed using a single instruction. Also processor speed is not the only factor in computing performance. To judge the performance of a computing system you need to evaluate disk and memory access time as well as processor speed. Please note also that the IBM System z9 that runs at 26 MIPS is slowed down artificially by microcode (idle cycles) to allow software that is priced according to the speed of the processor to be sold at a lower price than the full processor speed. -- Myself 14:15, 15 February 2007 (UTC)  —Preceding unsigned comment added by 194.11.209.148 (talk)


 * Yup. As the linked-above million instructions per second section of the instructions per second page says, "MIPS are not comparable between CPU architectures."  It also says "zMIPS refers to the MIPS measure used internally by IBM to rate its mainframe servers (zSeries and System z9), so if "26 MIPS" is really 26 zMIPS, those probably have nothing whatsoever to do with the "MIPS" in the "22 MIPS" for the Mac Quadra. Guy Harris (talk) 18:51, 15 February 2008 (UTC)


 * Yeah, there's a reason why MIPS is also traditionally an acronym for Meaningless Indicator of Processor Speed :-) RossPatterson (talk) 23:32, 15 February 2008 (UTC)


 * BTW, the 68020 in the Mac Quadra had more complex addressing modes than have, as far as I know, any S/360-to-System z processors - no S/360-to-Sz processor had, as far as I know, auto-increment or auto-decrement addressing modes. I don't know whether they picked up any of the scaled addressing modes, in which the index register was shifted left 1 or 2 bits before being added in.  So, whilst there are some complex instructions in S/360-to-Sz that were more complex than the most complex 68020 instruction, it's not as if the S/360-to-Sz instruction set is, across the board, more complex than the 68020 instruction set.


 * (Also, the impression I have is that the later S/390 processors, and the System z processors, implement simple instructions purely in hardware, without microcode, and implement more complicated instructions by internal traps to code in a special area of memory, with that code using a mixture of standard instructions and special and possibly processor-dependent instructions, similar to what the DEC Alpha did with PALcode. Are the artificial slowdowns implemented only in the millicode for the complex instructions, or is there also hardware that adds extra idle cycles that's turned on by CPU millicode on slower processors?) Guy Harris (talk) 23:55, 15 February 2008 (UTC)


 * Even within a single family the MIPS rate is meaningless. Compare the code sequences

MVC  FOO(256),BAR

MVC  FOO(1),BAR MVC  FOO+1(1),BAR+1 ...  ...              MVC   FOO+255(1),BAR+255


 * The first is much faster but has a much lower MIPS ratio. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:09, 30 August 2010 (UTC)

The million instructions per second page doesn't do a very good job of explaining it. For many machines in the early 1960's, the amount of processing done per instruction was about the same, so it might have made sense to compare actual MIPS. But it was soon realized that you couldn't really do that, so specific benchmark programs were used to calculate publishable MIPS. Different benchmarks were used for scientific and commercial system, and also they changed with time. Too often, someone comes along and tries to rate a machine based on actual instruction rate, possibly even comparing an eight bit processor to a 64 bit processor. Many of the benchmarks used won't run on smaller processors, adding more complication. Yes, give relative performance on different benchmark programs, but don't call them MIPS! Gah4 (talk) 21:30, 28 June 2016 (UTC)

Is this computer, "xyz", a mainframe ?
"Mainframe" might best be defined by example, list the 1st mainframes by the various manufactures. Were the early giants, ENIAC, AN/FSQ-7, ... mainframes? (if weight is the criteria, they're in!). —Preceding unsigned comment added by 69.106.232.37 (talk • contribs) 15:03, 15 October 2006


 * No, the Great Old Ones weren't mainframes. They were ... "computers". With cool names, like JOHNNIAC, SILLIAC, and GEORGE.  And with the exception of the family of IAS machines, they were one-offs. The "mainframe" era usually starts with the IBM 704 or 7090, and the commercialization of the UNIVAC. RossPatterson 03:11, 14 November 2006 (UTC)
 * The Soviet Strela computer (ЭВМ "Стрела") was not a one-off. There were at least 6 copies which were manufactured in the Soviet union. --Ancheta Wis 09:22, 1 August 2007 (UTC)

I think "mainframe" is defined very incorrectly in the main article where the following is stated, "Today in practice, the term usually refers to computers compatible with the IBM System/360 line, first introduced in 1965. (IBM System z9 is IBM's latest incarnation.) Otherwise, systems with similar functionality but not based on the IBM System/360 are referred to as "servers." However, "server" and "mainframe" are not synonymous (see client-server)". In my view, "mainframe" can be used as a term for all general purpose computers prior to the age of PCs and client/server servers. After PCs and servers came along, "mainframe" can be used for all of the other large systems. For sure, "mainframe" is not a System/360 based term and I have held long discussions about ICL and RCA mainframes over the years. {Bruce Palmer -- 8/29/07} —Preceding unsigned comment added by 75.68.166.182 (talk) 15:27, August 29, 2007 (UTC)


 * It might not be so far off. Machines like the PDP-10, CDC Cyber series, and the Cray-1 should be considered mainframes, but all are out of production. I believe Unisys still builds system that should qualify as mainframes, but they aren't well known outside of some government agencies. There were many smaller than mainframe systems produced in the 1970s and 1980s, usually called minicomputers, or even superminicomputers. In some cases, the replacement for mainframe systems is a cluster, possibly integrated into a large box, or smaller computers. I don't believe that those should be called mainframes, but they should have their own name. Others might disagree. Gah4 (talk) 21:39, 28 June 2016 (UTC)


 * I disagree. Systems such as the Bendix G-20, CDC 1604, Burroughs B5000, GE 625, Honeywell 800, IBM 7090, RCA 601 and UNIVAC 1107 were clearly mainframes, but systems such as the Bendix G-15, CDC 160, DEC PDP-5, Honeywell 200, Honeywell DDP 316, IBM 1401, RCA 301 and UNIVAC 1005 clearly were not, although they were general purpose. There was a thriving market for mini and midi computers both before and after the IBM S/360.


 * Machines such as the CDC 6600, ETA10, Cray-1, IBM 7030 and TI Advanced Scientific Computer were mainframes, but were more commonly referred to as supercomputers.


 * The Unisys systems descended from the Burroughs and UNIVAC mainframes are mainframes. Similarly the BULL machines derived from the GE mainframes. Shmuel (Seymour J.) Metz Username:Chatul (talk) 17:19, 30 June 2016 (UTC)


 * I was trying to answer, besides MOS:TENSE, what is a mainframe today, not what was a mainframe some years ago. Maybe that isn't what was being asked, but the descendants of IBM S/360, today's z/Systems, I believe are today considered mainframes that you can actually buy, and also some Unisys descendants of the machines mentioned above. Gah4 (talk) 20:41, 30 June 2016 (UTC)


 * It was It might not be so far off. thI was disagreeing with. of the BUL and UNISYS large systems that are stil in production should be called mainframes. ~


 * As far as I can tell, BULL mostly sells Intel Xeon based servers. But I do agree that the Unisys machines should count as mainframes. How many mainframes do BULL and Unisys sell, compared to IBM? They probably won't tell you.  Gah4 (talk) 23:30, 6 July 2016 (UTC)


 * I don't consider clusters of smaller computers, today's supercomputers, even when connected by high-speed I/O networks, to be mainframes, but others might disagree. I believe that they need their own category. Clusters of independent machines, such as the Amazon cloud, also need a new category, though the aggregate computing ability might be high, and they might even be mounted on a common iron frame. Gah4 (talk) 20:41, 30 June 2016 (UTC)

XDS
I'd like to clarify the role of XDS in history. XDS was the new name after Xerox purchased Scientific Data Systems, and SDS was influential at one time. I don't inow whether XDS includd the 940, 945 and 9300, or only the Sigma line. Shmuel (Seymour J.) Metz Username:Chatul (talk) 19:04, 25 August 2016 (UTC)


 * The SDS 940 was certainly an influential timesharing machine. Perhaps you would call it a mainframe. I'm not sure XDS ever marketed it, I see some documentation for it labeled XDS, maybe they provided legacy support. I would classify the Sigma systems as mainframes, maybe around a 360/50 and up off the top of my head. They were wonderful systems, but I don't think they ever got much traction, and I've heard it said that Max Palevsky sold Xerox a bill of goods when he unloaded SDS. The original edit I tried to fix mentioned PARC as significant, as indeed it was, but it had nothing to do with XDS, Sigma, or mainframes of any kind. I'm not sure XDS ever rose to the level of "Dwarf". Peter Flass (talk) 23:50, 25 August 2016 (UTC)


 * I'd certainly count the SDS 9300 as a mainframe; I'm not sure about the 940 and 945. I'd count the Sigma 7, am not sure about the Sigma 5 and would not count the Sigma 2. Shmuel (Seymour J.) Metz Username:Chatul (talk) 19:50, 2 September 2016 (UTC)

The following sentence does not make much sense:
"In addition, mainframes are more secure than other computer types: the NIST vulnerabilities database, US-CERT, rates traditional mainframes such as IBM zSeries, Unisys Dorado and Unisys Libra as among the most secure with vulnerabilities in the low single digits as compared with thousands for Windows, Unix, and Linux.[5]"

In particular, the phrase "... as compared with thousands for Window, Unix and Linux." This is comparing mainframes (hardware) with operating systems (software). The IBM zSeries run (or can run) Linux. — Preceding unsigned comment added by 75.147.4.177 (talk) 15:45, 13 November 2013 (UTC)


 * This may be difficult to express correctly and succinctly. One issue is that on System Z mainframes Linux is always run in a virtual environment and there are levels of protection below the operating system. --Boson (talk) 16:13, 13 November 2013 (UTC)


 * I think you can run zLinux on "bare metal", well, in a LPAR anyway, though I doubt many do, outside of Hercules systems.  The sentence is incorrect because (I assume) that what's being rated is a HW/SW combination, not just "mainframes'" but, e.g. zOS running on some platform.


 * Doesn't zLinux rely on CP for some services, e.g., recovery from hardware errors? Or am I confusing Linux with Solaris? Shmuel (Seymour J.) Metz Username:Chatul (talk) 21:24, 13 November 2013 (UTC)

This is a little complicated. I remember during the transition from SunOS to Solaris, and where many bugs in SunOS might have been exploited, but pretty early on the attacks were mainly on Solaris. At the time, Solaris was popular for web servers. But as for vulnerability, consider SQL injection. A web server front end (maybe on Windows, Solaris, or Linux) might connect to an SQL server on a mainframe. If the front end lets through SQL queries that it should block, it isn't the mainframe's fault. I have, often enough, had SQL errors returned from web sites that were just passing back what came from the SQL server. Very strange. Gah4 (talk) 01:10, 27 October 2016 (UTC)

Teletypes in the 1960's?
While use of a Teletype® was common in minicomputers, and while the later RCA Spectra 70 used a Teletype®, that was not a common device for mainframe consoles. Several machines used versions of an IBM Selectric Typewriter, but I'm not aware of any mainframe other than the Spectra 70 that used a Teletype®. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:53, 8 July 2012 (UTC)


 * I am not sure, but I suspect that the later models of Teletype, more modern than the ASR 33, would have made fine console devices. As far as mainframe and not console, I do remember using ASR 33's with WYLBUR, connected through a 3705. The hard part was finding the magic letter to tell the 3705 which baud rate you needed. Gah4 (talk) 00:58, 27 October 2016 (UTC)


 * OS/360 supported ASCII terminals (as opposed to consoles) early on, and I don't recall any special problems supporting faster Teletype models. However, mixing ASCII terminals with 1050, 2740 or 2741 on the same dial port was a problem. Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:04, 27 October 2016 (UTC)


 * As well as I know it, the 270x are what today would be called dumb. I suspect that the baud rates were hard wired, so even mixing different speed ASCII terminals would not work. I didn't know ASCII terminals connected to 360/370 systems until the 3705.  Also, I believe that the 270x shift the bits from the other end, so ASCII bits would be reversed in memory.  Nothing that TR wouldn't fix, though. Gah4 (talk) 20:28, 27 October 2016 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified one external link on Mainframe computer. 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/20150730182332/http://www.unm.edu/~tbeach/terms/types.html to http://www.unm.edu/~tbeach/terms/types.html

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

Cheers.— InternetArchiveBot  (Report bug) 17:08, 30 May 2017 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified one external link on Mainframe computer. 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/20111227021230/http://www.graph500.org/ to http://www.graph500.org/

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

Cheers.— InternetArchiveBot  (Report bug) 10:33, 13 January 2018 (UTC)

Supercomputer vs Mainframe
Is this correct? (One side effect is that even older software can benefit from adding mainframe CPs.) If so what does CPs mean? Or does it really mean CPUs?--Doc0tis 21:00, 23 November 2005 (UTC)


 * Yes, it's true. "CP" means "Central Processor", to distinguish it from the other processors that are often located on the slab of ceramic or silicon.  CPs run the instructions of a program, just like what most other architectures call a CPU.  Other processors do things like I/O handling, cryptographic primitives, or processing specialized workloads (databases, Java, etc.).  But "CP" is definitely a term of art, so it's history in the article. RossPatterson 03:00, 14 November 2006 (UTC)


 * Not really answering that question; there's a (much newer) banner on the relevant section, and I do not see much wrong with it (well, never heard of gameframe before, nor the game Taikodom; that seems to be dead, it's web page is..). Seems ok (maybe it was fixed); should we remove the banner. comp.arch (talk) 22:02, 26 October 2016 (UTC)


 * Not quite. CP is central processor, but it need not be implemented with integrated circuits and where IC technology is used, other processors need not be on the same chip. Further, the peripheral processors on, e.g., the CDC 6600 also execute instructions. Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:22, 2 December 2018 (UTC)

Parallel Sysplex is not virtualization
Parallel Sysplex is a means for separate OS images to cooperate, not a virtualization facility. The reference needs to make that clear. Note that Sysplex does simplify load shifting when installing OS upgrades, so the reference should be reworded rather than removed. Shmuel (Seymour J.) Metz Username:Chatul (talk) 19:09, 6 January 2019 (UTC)