Das Mailüfterl – A Vistit to the Vienna Technical Museum

The Mailüfterl (or, formally, Binär dezimaler Volltransistor-Rechenautomat – binary-decimal fully transistorized computing automaton) was one of the first fully transistorized computers and the first in continental Europe, built in Vienna by Heinz Zemanek et al. at TU Wien (technical university of Vienna) in 1956-1958. It was Austria's first real computer, a 48 bits machine designed to work in both decimal and binary systems. While it's a quite remarkable machine, documentation on it is rather scarce. So I'm providing some along with the pictures, mostly based on Zemanek's recounts.


History and Name

Having finished the URR1, a parallel relay computer (1950-1954, 18 bits word length, 16 words storage, multiplication, division done by a rotating switch), Heinz Zemanek was looking for another project to explore some more cutting edge technology. Starting on an initial grant of just 30,000 Austrian Shilling (1 ATS then converted to 26 USD, making this the equivalent of $ 1,154) Heinz Zemanek, who was well aware that he would need at least ten times the money for completing anything resembling a viable machine, began to look for some additional resources. A tour around European manufacturers in 1954 eventually secured pledges on transistors and diodes by Philips/Netherlands and other electronic components by Siemens. Notably, as transistors where just about to hit the market, the only ones available in suitable numbers were original designed for hearing aids and not optimized for fast switching.

When Heinz Zemanek introduced his project at the first German conference on computers in Darmstadt (GAMM/NTG Conference in October 1955 [1]), Zemanek concluded, alluding to the bold names of US-American machines and the supposedly slow performance of his hearing-aids-parts, "In other words, it will not be a typhoon, whirlwind, or hurricane, but a nice Viennese spring breeze would be sufficient for our purposes."[2] – Actually, he didn't say spring breeze, and he didn't say May breeze either, but "Mailüfterl", a special Austrian/Viennese diminutive form, traditionally providing the title of a few waltzes. Overpowered by this amount of Viennese flair Zemanek's German colleagues seriously urged him to stick with the name – and so he did, even against some opposition. (The Austrian chancellor later remarked bitterly in his memoirs, "Such a serious endeavor and such a ridiculous name!")

[1] GAMM-NTG-Fachtagung. Elektronische Rechenmaschinen und Informationsverarbeitung. Electronic digital computers and information processing. Darmstadt, October 1955.

[2] Oral history interview with Heinz Zemanek (Charles Babbage Institute, 1987):
http://conservancy.umn.edu/handle/11299/107723

Note: Alternatively recollected by Zemanek as, "Also, Wirbelwind wer ma keinen derbauen aber auf ein Wiener Mailüfterl sollte es reichen", and, "Auf einen Taifun oder einen Wirbelwind wird's nicht reichen, aber auf ein Wiener Mailüfterl schon."


The Project

The built of the Mailüfterl was, appropriately for its name, started in May 1956 and was completed in May 1958 with its first run on May 27 (calculating the prime 5,073,548 261). The project was famously conducted as a semi-illegal one, as Zemanek held just the position of an assistant on his own with no boss above him, since the sede was vacant. At one point, the project was actually killed as Zemanek didn't have the blessing of at least 3 professors, required for a project of this size, and wasn't able to produce such either. But Zemanek refused to do the paperwork and, especially, writing all those letters of cancellation to the various companies that had committed to donating to the project, thus leaving this to the ones who had halted the project. But these also backed off of the inherent horrors of bureaucracy and the project carried on, giving the perfect example of what is called an "Austrian solution".

Eventually, the project completed on a total funding of ATS 250,000 (USD 9,615) as well as donations by Philips (3,000 transistors and 5,000 diodes), Siemens (20,000 capacitors and resistors), and help on the custom drum by Elin/Austria.

The completion of the software continued until 1961, when Zemanek, his team, and the machine were transferred to IBM Austria, where the machine was used for a vocoder project that eventually became the vocoder for the System/360. (Zemanek's division, the IBM Vienna Laboratory, was most influential for the invention of the Vienna Definition Language [VDL, 1964-68, https://en.wikipedia.org/wiki/Vienna_Development_Method], ultimately linking Wittgenstein to the definition of PL/I.)

The machine was decommissioned and put to storage in 1966, allegedly for wear of the drum's surface coating. In 1973 the machine was transferred to the Vienna Technical Museum (Technisches Museum Wien).

The machine is built entirely of chip cards/modules and very thin (may we call this the first tablet computer?). It is now on display in a somewhat thicker, protective housing than the original frame and has lost its bold display of the machine's name (compare: https://www.wien.gv.at/ma53/rkfoto/2009/1973g.jpg).


Software

The machine was designed with an eye on high level languages, originally intended for use with Fortran, but Zemanek wasn't lucky on this as for any support by IBM International. So he eventually settled for Algol (the first book in German on Algol 58 was written by Heinz Zemanek) and later LOGALGOL.

One of the more glamorous use cases was also the first real application, answering a request by a Viennese composer: In twelve-tone music, there are two kinds of twelve-tone series providing the very building-blocks: A series may be either constructed of each key of an octave once or by using each interval once. But, how many series are there fulfilling both of the conditions? Eventually Mailüfterl provided the answer after a 60 hours run. (For those curious: The result is 2 × 1928 – compare: Nachrichtentechnische Fachberichte 4, 1956, p. 56-59.)


Technical Data

– Built May 1956 – May 1958
– binary and decimal modes (per instruction using a mode flag)
– 48-bit words accessible in groups of 4 bits
– 133 kilocycles (133 kHz)
– multiplication/division by an internal 5 steps program
– 3,000 transistors, 5,000 diodes, and about 20,000 capacitors and resistors on 1,000 chip cards
– about 30 km of wires
– drum storage: 200 tracks, 50 48-bits words each, 4 bits intersection, average access time: 10 milliseconds
– core memory: 50 48-bits words (1 drum track) and additional 50 directly addressible registers in I/O circuitry.
– in decimal mode numbers were apparently implemented in excess-3-encoding (also known as XS-3: add 3 to get the binary representation, like early Univacs)
– total budget ATS 250,000 (USD 9,615)[3] + donations by Philips and Siemens
– total runtime (1958–1966): 12,000 ... 15,000 hours.

Most of the circuits had been designed for 250 kHz, but a few weren't and 133,000 steps per second were agreed on as a safe speed for security and reliability of operations. (Not too bad for the hearing-aids transistors, when comparing this to the 200 kilocycles of the commercial DEC PDP-1 introduced in 1959.)
The ferrites used for core memory weren't of the best quality and had serious reliability issues, resulting in a patent on a more reliable method of accessing the cores, later also used in Zuse machines. (Konrad Zuse, a close friend of Zemanek, sponsored the wages of the team's architect in chief on the condition of him transferring to Zuse after finishing the project.) Legend has it that the core memory was hand-knitted by a Viennese lady.

The theoretical address range was 10,000 words (4 × 4 bits address component accessing 10^4 registers).

A remarkable piece of special I/O equipment was an electronic chess board, used for a demonstration program solving a special endgame situation (2 kings and one rook, according to Heinz Zemanek, it never worked reliably).

[3] To put the budget of US$ 9,615 (ATS 300,000) into proportion: The price of an average new car in the US was ca $ 2,000 in the mid-1950s.


Instruction Code

The machine used a quite flixible system of instructions, called minima principle or Freiburg code, somewhat related to microprogramming (compare also the ZEBRA machine by van der Poel):

4 bits (1 bit used) flag binary/decimal
4 bits conditional execution codes (15 conditions)
4 bits main instruction code (15 codes)
3 bits auxilary instruction code (7 codes)
9 functional bits (e.g., flags for storage select)
2 × 4 bits address index (10^2)
4 × 4 bits address (10^4)


Team

Heinz Zemanek (project lead)
Kurt Bandat (core store)
Rudolf Bodo (layout/architecture)
Viktor Kudielka (drum store)
Kurt Walk (transistor physics)
Peter Lucas (programming, joined later)
Eugen Mühldorf (tape reader, part time)


Related Links

* Google image video (watch this, although in German): https://www.youtube.com/watch?v=lvaiEW-Sm4A
* TV portrait (1995, German): https://www.youtube.com/watch?v=vdF3Z3jnk4Q
* Heinz Zemanek as a beloved talker: https://www.youtube.com/watch?v=amGw_g2iZ7k
* audio only: https://www.youtube.com/watch?v=L08gnO4MAyQ

Website: http://www.zemanek.at

Mailüfterl Emulator (Java): http://members.aon.at/~nkehrer/mailuefterl.html

Technisches Museum Wien: http://www.technischesmuseum.at
PhotoPhotoPhotoPhotoPhoto
2016-06-30
33 Photos - View album
Shared publiclyView activity