Manchester Mark I

:"This article is about the early British computer"." The term "Manchester Mark I" can also refer to the Avro Manchester heavy bomber in RAF service during the early stages of World War II".

The Manchester Mark I was one of the earliest electronic computers and the first electronic stored program computer, built at the University of Manchester in England, in 1949. It was also called Manchester Automatic Digital Machine, or MADM. It was developed from the Small-Scale Experimental Machine (SSEM) or "Baby". It is especially historically significant due to its pioneering inclusion of a kind of index register in its architecture, as well as being the platform on which Autocode was developed, one of the first "high-level" computer languages.

Development of the Mark I started after the SSEM demonstrated the utility of the stored-program approach, which dramatically improved a machine's flexibility. This approach was being looked at by other researchers, notably Alan Turing's efforts on the Pilot ACE, Cambridge University's EDSAC, and EDVAC in the US. The SSEM differed primarily in the choice of memory system, using the much faster Williams tubes instead of mercury delay lines.

With the successful demonstration of the SSEM, the British government contracted Ferranti in October 1948 to deliver a full-scale machine based on its basic concepts. Key improvements in the design were going to include a magnetic drum for loading programs into the machine's Williams tube memory, replacing the SSEM's paper tape, the addition of index registers and a hardware multiplier. The word size was increased slightly from 32-bits to 40, read and written as four 10-bit "short words". Instructions used a single short word, addresses two, and numeric data four. Although the 10-bit instructions could hold up to 1,024 different codes, the machine only had 30 in its final version. Standard instruction time was 1,800 microseconds, but multiplication was much slower. The Ferranti Mark I (based on the Manchester Mark 1) had an addition time of 1,200 microseconds and a multiplication time of 2,160 microseconds.

The SSEM included two registers on its Williams tube, the accumulator A and program counter C. Mark I added another, D, for holding one side of a multiplication, leaving B the natural place to hold the index register. Since the system used a 20-bit address, the "B-line" on the tube held two address offsets. This is the earliest known implementation of such index/base registers – an important innovation in computer architecture, unknown in other machines until the emergence of second-generation computers (approximately 1955–1964). The Mark I included two tubes, each storing 64 rows ("double density") of 40 points, for a total of 128 words. 64 words was considered to be a single "page", so the system stored 4 pages. Freddie Williams deliberately sized the drum to store two "pages" of Williams tube data – that is, 2x32x40 = 2,560 bits – per track, and 32 tracks in total. The drum was timed to spin at the "refresh" rate of the Willams tubes, allowing pages to be read and written between refreshes, a task that took about 30 cycles.

The first version of the machine was running in April 1949, known as the "Intermediary Version". This version was largely feature complete, but lacked input/output instructions to move data from the drum to the tubes or paper tape to the drum.

The first realistic program to be run on the Mark I was a test of Mersenne primes, run in early April 1949. The computer ran error-free for 9 hours on the night of June 16-17, 1949. The "Final Specification" version was completed in October 1949, adding a second drum and various instructions to read one line of data to and from the drum to tubes and drum to paper tape. Over time the existing drums were used to store more data, typically 47 tracks.

The machine used 4,200 vacuum tubes for logic, which proved to be a terrible reliability problem. In one calculation the machine spent almost 25% of its time "down", due both to the tubes and the drums. Nevertheless the University was successful in attracting commercial users to rent time on the machine for £50 an hour.

After the Mark I was running, development continued in several directions. Dick Grimsdale and Doug Webb attempted to improve the reliability of the Mark I by building the machine out of transistors, perhaps being the first transistorized computer when their prototype ran in November 1953. Their work was later picked up by Metropolitan-Vickers to create the Metrovick 950, of which seven were sold.

The main Mark I team, Tom Kilburn and Freddie Williams, concluded that computers would be used more in scientific roles than pure math, and decided to start development of a new model including a floating point unit. The resulting machine, Meg, was both simpler than the Mark I as well as much faster for math problems. Ferranti, who had built the Mark I, rebuilt Meg with core memory and sold the resulting design as the Ferranti Mercury.

Among the Mark I team were mathematicians Conway Berners-Lee and Mary Lee Woods, who would later marry; their son, Tim Berners-Lee, is acknowledged as the inventor of the World Wide Web.

The Met.Office used the Mark I to develop its first prototype computer programs for weather forecasting; they used difference equations to calculate atmospheric pressure.

References

*cite book
last = Lavington
first = Simon H.
title = Early British Computers
publisher = Manchester University Press
date = 1980
isbn = 0-932376-08-8

External links

* [http://www.computer50.org/mark1/MM1.html The Manchester Mark I]
* [http://www.cs.man.ac.uk/CCS/res/res04.htm#g Early computers at Manchester University] in RESURRECTION The Bulletin of the Computer Conservation Society ISSN 0958-7403 Volume 1 Number 4 Summer 1992
* [http://alpha60.de/research/muc/ A simulator of the Manchester Mark I, executing Christopher Strachey's Love letter algorithm from 1952]
* [http://doi.ieeecomputersociety.org/10.1109/85.222841 cite journal
last = Lavington
first = Simon H.
title = Manchester Computer Architectures, 1948-1975
journal = IEEE Annals of the History of Computing
volume = 15
issue = 3
pages = 44–54
publisher = IEEE
date = Jul-Sept 1993
doi = 10.1109/85.222841]