Hacker News new | past | comments | ask | show | jobs | submit login
Mainframes and Supercomputers, From the Beginning Till Today (cpushack.com)
85 points by stargrave 11 months ago | hide | past | web | favorite | 27 comments

> The first mainframe was created by the well-known IBM in 1964

Were the IBM 700/7000 series (1953-1964) not mainframes? I would say they were.

And what about other vendors pre-1964 machines, e.g. UNIVACs. Were they not mainframes too?

> you need to determine what the super-computer differs from the mainframe and which is faster

I'm not convinced that "mainframe" and "supercomputer" are necessarily mutually exclusive categories. What about the IBM 7030 Stretch? Doesn't it belong to both?

There have been supercomputers with an S/360-descended architecture. For example, the Fujitsu FACOM VP and VP2200 series, the Hitachi HITAC S-810. Such machines were both supercomputers and IBM-compatible mainframes, and could run variants of MVS.

A good way to distinguish a mainframe from a supercomputer would be served with... a car analogy.

A race car is often uncomfortable and hard to drive. You need to be physically fit to drive it. Supercomputers are the race cars of computers. You need to do crazy tricks to extract their maximum theoretical performance. Mainframes, however, are more suitable to run common tasks, are easier to drive and generally more reliable. A mainframe like a 360/195 would be somewhat like a BMW M-series. Its performance is impressive, but you don't need to practice your whole life not to kill yourself in one.

The problem with the "car analogy" is that some supercomputers are also mainframes.

For example, look at the 1st Top 500 List, from June 1993. In position 36, was the Hitachi S-3800/480 at the University of Tokyo [1][2]. The S-3800/480 was an S/370-compatible machine with vector processing extensions. It supported two operating systems, OSF/1 Unix and also VOS3, which is a fork of IBM MVS [3]. Arguably, if a machine is S/370-compatible and runs a derivative of IBM MVS, it is an IBM-compatible mainframe. Yet, also, it was a supercomputer. It could run the same business and scientific software as any other IBM-compatible mainframe. But scientific code can take advantage of the special vector processing instructions (either through assembly or through the Fortran compiler.)

This was probably the very tail end of this category of machines, IBM(-compatible) mainframe supercomputers. But if you go back to the 1980s and earlier, it was actually a quite common category, with instances being supplied by IBM, Amdahl, Fujitsu and Hitachi. (Unfortunately, the Top500 list only goes back to 1993, so it omits most of this history.)

[1] https://www.top500.org/list/1993/06/?page=1

[2] https://www.top500.org/system/168780

[3] https://en.wikipedia.org/wiki/MVS#Closely_related_operating_...

> The S-3800/480 was an S/370-compatible machine with vector processing extensions.

IBM also had the Vector Facility for its 3090 series. I'd still say these are the M-series (or the post-Audi Lamborghinis) of computers - fast, but they don't require you to sacrifice comfort for speed.

Very cool.

I wish I was able to locate a Thinking Machines CM-5 before they were all destroyed. Even the front indicator panel would have been nice.

I’ve created my own smaller version with an Arduino and some LED panels but its not the same.

I'm thinking about building a panel that'd would gather core activity of a couple machines and map that onto one or more 8x32 LED matrices. The core activity would be gathered from a client running on the machines themselves and sent to the machine running the panel.

On my tests running ffmpeg, however, the results would be disappointing: all LEDs light up at the same time...

Disappointing it does not mention Convex Computer, which:

-produced the world's first commercially available gallium arsenide supercomputer, the Convex C3

-owned the minisupercomputer market from ~1986 through its end in 1995

-built the first commercially successful flat address space, cache coherent massively parallel processor the SPP1000 series in '94

I agree wholeheartedly :-) Incidentally, I have 9 Convexes at home: https://vaxbarn.com/index.php/43-project/641-the-convex-comp...

-shipped an auto-vectorizing C compiler instead of emphasizing just Fortran

-hired Tom Christiansen and shipped with Perl pre-installed

-employed Robert Morris (a co-founder of Y Combinator) as a summer intern

> During the first month of IBM, orders were received for more than a thousand copies of such machines, and in the 6 years of the family’s existence, more than 33,000 such machines were sold.

What for? I can imagine some tasks for such a machine in the middle of XX century, but not for thousands of them.

The IBM system/360 wasn't exactly a super computer, but a modular, scalable family of mainframes, ranging from medium business machines to scientific computers. The S/360 built on previous IBM mainframes, like the IBM 605 (first mass produced computer), the IBM 305 RAMAC (hard disk!), the IBM 7070 (solid state) and the popular IBM 1404 (about 10,000 units sold, an image of it found its way amongst the photos of the S/360 on that page). The S/360 was especially important for introducing the octet for a byte.

It may be noted that the IBM 1401 was introduced in 1959 to replace hardwired punched card appliances, especially after these were heavily challenged by the Bull Gamma 3. The lower range IBM S/360 were replacing the 1401 in turn. The S/360 family was eventually replaced by its direct successor, the IBM system/370.

That said, IBM did maintain some real super computing projects around this time, namely project X and project Y (resulting in the ACS project). Project X eventually became the top of the range S/360 Model 92, project Y (ACS-1) was especially important for its legacy in cache architecture and software optimization (optimizing compilers, etc.)

Edit: Project X was actually sold under the designations S/360 Model 91, Model 95 and Model 195. (Introduced in 1969, these were more an add-on to the family, which had been introduced in 1965 and was already approaching the end of its retail cycle.) About 3 dozen of these actual super computers were built in total[1].

[1] https://pdfs.semanticscholar.org/7726/d267b6fb7470066640e7bb...

The greatest number of those sold and installed were for mundane business data processing. By the mid-60s every medium and large business needed a computer. The slowest 360, the model 30, would have had trouble competing with a Z80 of 20 years later. However, the 360/91 was genuinely a supercomputer, and it pioneered out-of-order execution among other performance features.



The SAC air defense system for one. Also this was the age of giant industrial enterprises. These things were used for payroll, etc.

As I recall, it was widely believed that there was an economy of scale for mainframe computers, so you were better off serving all of your users with a single timesharing machine if possible. If most of your programs ran in batch mode, then you might need 64k, but only for a couple seconds a day, so you shared those 64k with 1000 other people.

It’s evident, scrolling through that, how drab computers have become. The 360 had a giant control panel with blinken lights, and cool looking tape drives. The Cray looks cool, despite the lack of control panel. Everything else is just a drab rack.

I just missed the 360 era when I became an operator, but the 370 console was similar in sci-fi-ish feel. There was a hexadecimal rotary dial via which you'd set the boot device address, and an actual mechanical switch to boot it up.

In contrast, the 3033 operator console was just sort of a glorified terminal.

I thought the Quiesce feature was so cool -- you could freeze all programs and OS activity in place, then disconnect and work on any devices or peripherals at will, then resume.

In a mere 2-year span, I got to run 370s, 3033s, 308x (including the 4-processor 3084), and 309x not long after I left operations for programming.

The PDP 8 and 11 and cool front panels.

I recall on our PDP 11/40 if the separate module that handled the boot failed, you had to toggle the boot loader in by hand using OCD (octal coded decimal) which was a pain.

Much of that has with miniaturization.

Those massive tapes had downright abysmal storage capacities by today's standards.

And binkenlights vanished when ICs could switch so fast that the human eye would just see a steady light (much like we can't see a fluorescent lamp blinking).

Those lights were directly connected to the main bus which ran at several hundred khz to the low mhz. You never looked at them while the machine was running except as a way to confirm "something" was happening. You would not see individual pulses but rather they would appear dim or bright depending on activity.

Instead the machines had a single step mode where you could force the computer to execute one instruction at a time. Or you could pause the machine completely and manually poke bits into memory yourself.

The real death of the front panel was when machines got their own bootloaders and could load their operating system without the operator manually injecting instructions on how to read from external storage directly into memory. This actually happened much later than you might appreciate. This combined with software based debuggers which could be more powerful.

I'd agree, with the availability of glass monitors for diagnostic purpose and the end of switches for toggling in initial loaders blinkenlights became obsolete.

The transition is probably well illustrated by the Datapoint 2200 smart terminal (which famously provided the blueprint for the Intel 8008): An early illustration shows the machine with a DEC mini style row of console switches [1], but when it was announced in 1970, the toggle switches were gone. Ironically, the early 8-bit micros which sprang off from the DP 2200 were probably amongst the last machines to come with toggle switches and console lights (again, because of the lack of built-in diagnostic facilities).

[1] Concept illustration of the Datapoint 2200 (John Frassanito, 1969) – arguably, console switches were just what you expected from a "self-respecting" computer at the time, but, by the time the design had progressed from concept to actual hardware, these just didn't make any sense anymore: http://4.bp.blogspot.com/-2urGdLzMYf8/VnY76lBDUlI/AAAAAAAANS...

(From http://bugbookmuseum.blogspot.co.at/2016/01/datapoint-2200-8... – Mind that Datapoint is then still CTC and the terminal is marked "Veripoint 2200" in the concept illustration.)

I wish my new desktops looked as cool as the ones SGI made back in the day.

The Crays look futuristic even by today's standards. Having a supercomputer that could kill you with poisonous gas is the epitome of futuristic.

This is kind of a scatterbrained article, it doesn't tie in mainframes in a meaningful way. The CDC 6600 and IBM 7030 Strech would have been a better starting point for supers. The S/360 mainframe was primarily a business line of computers, although the 360/91 was a capable super.

So to expand on mainframes in particular, which are less well documented/understood by my generation:

The IBM mainframe continued being primarily a business computer but they often had enhanced capability models like the 3090 with optional vector facilities that put them in competition with dedicated supercomputers.

Few people that came into computing in the 1990s and thereafter realize that a lot of what we might assume to be "modern" capabilities existed since the 1970s, via these machines. The machines gave companies inter-office memos (email), document preparation (word) and document/image storage (like modern dropbox, gdrive etc), OLTP (like later RDBMS oracle etc), OLAP (like modern hadoop, terradata etc), custom applications and hardware for all kinds of line of business activities like automated logistics via bar code systems, point of sales cash registers and bar code scanners, ATMs, complex billing generators printers that could customize/cut/collate bills and notices, bill processors with check printers/OCR scanners, phone switchboard integration for call routing and auto-attendant etc.

VTAM and SNA allowed the machines to communicate, and intercommunicate with machines of other companies across the globe. Time sharing services were offered for smaller companies in a way not dissimilar to modern "cloud computing". International networks existed to interchange data between different computer types like https://en.wikipedia.org/wiki/Tymnet.

The machines were also critical to the design and engineering of large construction, manufacturing, PLM, and simulation. Aircraft, ship building, power plant design, space exploration, automotive, circuit board and VLSI chip design/layout/EDA.. all that was primarily done on these machines up until the early '90s when UNIX workstations took over. CADAM, CATIA etc lots of good history covered on http://mbinfo.mbdesign.net/CAD-History.htm.

It's actually kind of astounding how much these did and how quickly it was washed out of common knowledge. If you had a white collar job in the '70s-early '90s you probably directly and indirectly spent a lot of time on these machines. 3270 terminals and later PCs with 3270 emulation.

Today, the mainframe is still integral to the running of Western civilization, although they are almost exclusively back office transaction processing and batch reporting systems.

The table at the end of the article, showing roughly Teraflops as a function of year. I'd be interested to see a corresponding table for minicomputers, microcomputers and other desktop machines such as GPU's.

I'm sorry, but an article claiming to trace the history of supercomputers that doesn't even mention the CDC 6600 can't be taken seriously.

Also missing mention of the Cyber 205 and the Illiac IV.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact