issue number 90 summer 2020
TRANSCRIPT
Computer Conservation Society
Aims and objectives The Computer Conservation Society (CCS) is a co-operative venture between BCS,
The Chartered Institute for IT; the Science Museum of London; and the Science and
Industry Museum (SIM) in Manchester.
The CCS was constituted in September 1989 as a Specialist Group of the British
Computer Society. It is thus covered by the Royal Charter and charitable status of
BCS.
The aims of the CCS are: To promote the conservation of historic computers and to identify existing
computers which may need to be archived in the future,
To develop awareness of the importance of historic computers,
To develop expertise in the conservation and restoration of historic computers,
To represent the interests of Computer Conservation Society members with
other bodies,
To promote the study of historic computers, their use and the history of the
computer industry,
To publish information of relevance to these objectives for the information of
Computer Conservation Society members and the wider public.
Membership is open to anyone interested in computer conservation and the history of
computing.
The CCS is funded and supported by voluntary subscriptions from members, a grant
from BCS, fees from corporate membership, donations and by the free use of the
facilities of our founding museums. Some charges may be made for publications and
attendance at seminars and conferences.
There are a number of active projects on specific computer restorations and early
computer technologies and software. Younger people are especially encouraged to
take part in order to achieve skills transfer.
The CCS also enjoys a close relationship with the National Museum of Computing.
Resurrection
The Journal of the
Computer Conservation Society ISSN 0958-7403
Number 90
Summer 2020
Contents
Society Activity 2
News Round-Up 13
Packet Switching and the NPL Network
Peter Wilkinson
14
Thoughts on the ICL Basic Language Machine
Virgillio Pasquali
23
Obituary: Margaret Sale Stephen Fleming
25
Why is there no well-known Swiss IT industry? Herbert Bruderer
26
Ho Hum, Nothing to Do Michael Clarke
29
50 Years Ago ….From the Pages of Computer Weekly Brian Aldous
31
Forthcoming Events 35
2 Resurrection 90 Summer 2020
Society Activity
IBM Museum — Peter Short
Current Activities
The first couple of months this year were relatively quiet, with three
curators taking vacation, two of who got stranded in Australia and New
Zealand during March. Hursley closed for the duration at the start of March,
with employees working from home, so we’ve not been back there since.
Activities have been limited to website updates and thoughts for the future.
A new page has been added to the
website, which tries to explain the
typical processes and hardware
involved in data processing using
punched cards
slx‑online.biz/hursley/pcprocess.asp.
It also highlights some of the
hardware that in part contributed to
the demise of the IBM Card. It
includes a photo of a warehouse facility containing an estimated 4Gb of
punch card storage. Each punch card offered a maximum of 80 columns,
or bytes. One card box typically contained 2,000 cards for a maximum of
160K bytes, 6¼ boxes per Megabyte, or 625 boxes per 100Mb. Find your
little processing job amongst that lot!
We had an unusual enquiry from a lady in Ireland, whose father used to
run a typewriter rental business. Her parents gave her the middle name
“Pifont” after a Selectric typewriter golf ball that contained the π symbol,
and she in turn has given her daughter that middle name. She wanted to
show her daughter a picture of the golf ball and wondered if we could help.
After digging through our collection I managed to find one such type-head
and was able to send her a photo and a print-out of the complete Symbol
font set.
Resurrection 90 Summer 2020 3
The follow on plan was to see if we had a duplicate type head which we
could send, but the dreaded virus thwarted that plan. Meanwhile we found
a 3D printer file which I was able to update with a Greek font. My son ran
this through his 3D printer, but as expected the quality was nowhere near
good enough. We do have access to a resin printer in Hursley, so some
day we should be able to produce this in a much higher quality and
surprise Ms. Pifont.
Other
Last time I reported on the offer of a working System/3. The owner
subsequently informed us that he’s not yet ready to part with it, and we
may have to wait for illness or his demise. At least that gives us time to
try and find £5K for shipping charges. We live in hope!
We’ve also added a link from the museum website to IBM’s interactive
“History of Progress” at www.ibm.com/ibm/history/interactive/index.html.
This runs in your browser, windowed or full screen, using Adobe Flash
Player, and gives a really interesting account of IBM’s history from pre-IBM
1890s through to the early 2000s.There’s also a desktop version and a
PDF of the story to download.
Analytical Engine — Doron Swade
The Babbage technical archive held by the Science Museum has been
reviewed sheet by sheet so the reference sources of AE-related content
are now known. After a short hiatus Tim Robinson is carrying out the same
exercise on the Babbage material in the Buxton papers held in Oxford.
This material is of particular interest, not least because there are several
essays Babbage wrote on the Analytical Engine while in Italy immediately
after his visit to Turin in 1840 where he gave his first and only seminar-
lecture on the Analytical Engine at a convention of mathematicians,
surveyors and scientists. This rare engagement with others was a
significant stimulus to Babbage so his writings immediately following this
are of special interest. We have done two substantial photo shoots of this
manuscript material, so digitised images are to hand. Part of the difficulty
with is that the manuscripts are unsympathetically bound (text lost in the
binding gutters), some material is undated, and the manuscripts are not
bound in chronological order. We are also currently planning on the best
way to document the findings so far, for wider dissemination — this a
lesson learned from the material left by the late Allan Bromley who
regrettably published only a small part of his deep understanding of the AE
design.
4 Resurrection 90 Summer 2020
EDSAC — Andrew Herbert
For the past several weeks EDSAC has been successfully running our initial
test program — a simple sequence of no-ops followed by a jump back to
the beginning. This has confirmed correct operation of the instruction fetch,
decode and end pulse parts of the main control state machine, the correct
operation of the sequence control register, correct handshaking between
main control and the arithmetic unit and correct operation of the main
store coincidence and addressing system.
Separately the nickel delay line store and initial orders have been
commissioned as sub-systems but are yet to be connected to the rest of
the machine. (We have been using “silicon delay lines” during main control
commissioning to simplify the task of fault finding). A video will shortly be
uploaded to the project website showing the construction and
commissioning of the store and a demonstration of the initial orders unit
loading the initial orders into the nickel delay line store with a visual
representation on the operator’s display screens.
Having at long last seen a simple program execute reliably for many
minutes, the main control team has gone back and tidied up the circuits
and documentation to provide a base line for the next stage, namely the
commissioning of orders that either read operands from or write operands
to store (which is the majority of the order code). In addition James Barr
is embarking upon a redesign of the transfer unit which deals with aligning
operands for either short or long number arithmetic. (John Pratt designed
and build an initial transfer unit, but we have subsequently found the unit
needs to delay by at least one minor cycle in all cases and its re-alignment
is required by one and a half minor cycles — hence the need to revisit the
design.)
Assuming the team is not too disrupted in the coming months we hope to
make progress commissioning all of the arithmetical orders. James and his
team are confident they are decoding and signalling to the arithmetic unit
correctly and Nigel Bennée is confident the arithmetic unit calculates
correctly so we hope, once operand reading and writing is working, we can
progress to running useful programs.
The tape reader and circuits for input are ready for testing and will follow
checking out the arithmetic unit. The printer and circuits for output are in
construction.
There is beginning to be an ever brighter light at the end of the tunnel.
Resurrection 90 Summer 2020 5
Data Recovery — Delwyn Holroyd
We have made some progress on the road towards being able to read LEO
magnetic tapes. The ICT 1301 tape decks are based around an Ampex
TM‑4 transport, the same type used with LEO machines albeit with a
different head. However as I reported last year, we now have an 8‑track
LEO head.
The first step is a working deck and to that end Rod Brown and I spent a
day in TNMoC’s storage facility examining the deck we intend to use,
which according to Rod was the most reliable one when the ICT 1301 was
in use.
By the end of the day we had applied power and managed to feed tape
forwards. Unfortunately backwards was not possible as the actuator is out
of adjustment. Several other mechanical issues were evident and Rod has
taken away various parts to work on.
We had been hoping to have another go at it in April, but of course world
events intervened. We did however manage to move the deck from
storage to TNMoC just before the lockdown commenced. This will make it
much easier to work on the deck when we are able to.
Meanwhile another group has rescued an IBM System 360 Model 20 — you
can read all about it here: ibms360.co.uk . We've been asked to attempt
data recovery from a set of about 20 reels of tape that were found with it.
For this work I used our HP data recovery deck that is fitted with a
combined 7 and 9 track tape head.
I managed to capture raw data from all the tapes on my last visit to the
museum prior to the lockdown. Analysis is ongoing but it appears that all
the tapes have been overwritten with blank header labels. In some cases
I’ve been able to extract previously recorded data beyond the blank
header. This consists of lists of names and addresses of what appear to be
plumbing and sanitary engineering companies in Switzerland! There are
also a few dates in the 1980s. So unfortunately it appears that the tapes
were not used with the System 360, and do not contain the hoped for
software.
Turing Bombe — John Harper
We are rather concerned that, even if TNMoC was allowed to open with
some limitations in the future, our operators might not wish or would not
be able to give demonstrations whilst at the same time keeping a social
distance.
6 Resurrection 90 Summer 2020
ICL 2966 — Delwyn Holroyd
The failed 5V/75A store power supply mentioned in the last report has
been replaced with the spare and the failed unit cleaned and repaired. The
fault was an open circuit relay coil — the relay is part of the anti-surge
circuit.
The system console monitor and the 7501 terminal both suffered from
intermittent loss of horizontal hold, and the remedy in each case was
cleaning the adjustment pot.
Visitors to TNMOC will be aware that one of the programs used to
demonstrate the machine is Conway’s Game of Life, running under
Maximop. Sadly John Conway died on April 11th. XKCD published a cartoon
featuring a Life pattern of a figure that decays into a glider
(xkcd.com/2293). I’ve added this to the repertoire of our demonstrations,
ready to be installed when the museum is able to open once again.
The National Museum of Computing — Kevin Murrell
The museum is closed to visitors during the Covid-19 lockdown. We are
not expecting to open until the end of May at the very earliest, but it will
take a long time to build the visitor numbers back to the point before the
lockdown began.
We have put two of the full-time staff and all the casual staff on furlough
until the end of May, which may yet be extended. Our museum director,
Jacqui Garrad, is not on furlough and has worked tirelessly keeping the
museum looked after.
Andrew, Jacqui and I are pursuing all the funding opportunities available
and are beginning to have some success in raising new funding. Jacqui is
keeping the flow of information and entertainment from the museum going
via social media.
Our hugely valued volunteers have been working remotely and supporting
the museum as best as they can from home. I think we can expect many
new and interesting exhibits to start to appear as ‘soon as this is all over’.
Elliott 803 & 903 — Terry Froggatt
As you may remember, last year TNMoC was loaned an Elliott 920M by the
Rochester Avionic Archives (one of three that they have), and you may
wonder why I’ve not reported about it since. In fact there has been little
activity on this 920M, and considerable activity on other Elliott 920Ms,
which would usually fall outside my reporting brief. But the matters are
related, and at present there is little to report from TNMoC whilst it is in
lockdown, so here is the wider story.
Resurrection 90 Summer 2020 7
Firstly a recap. Anyone who has been to TNMoC will be familiar with their
Elliott 903 “011y” (that I routinely report on), which is one of a handful of
903s still working in the UK. This is a mid-1960s desk-sized machine,
many of which were used by research institutes and which later found
their way into schools. The CPU is essentially identical to the military 920B
which was used in Forward Artillery Control Equipment (FACE) and in the
Nimrod Mk 1 submarine hunting aircraft. The 920M is a shoebox-sized
equivalent, which was used from the late 1960s in the Inertial Guidance
System of the ELDO Europa Satellite Launcher, and in the Navigation &
Weapon Aiming Subsystem (NAVWASS) of the Sepecat Jaguar aircraft. I
was a member of the Jaguar software team, and we initially had access to
a 920M MCM2 with a 5μsec store, but we moved onto a 920M MCM5 with a
2μsec store as the code grew and for the production run of some 200
aircraft.
The 903 or 920B was built from discrete components, many of which are
mounted on matchbox-sized Logic Sub-Assemblies (LSAs), on a total of 76
boards measuring some 8” by 5”, which are all easily removed from their
rack for diagnostic card swapping. Spares for almost all of the components
are available, and we have a full set of engineering manuals and circuit
diagrams. The 920M is a different kettle of fish. It is approximately 32lb of
Araldite, holding some electronics in tight formation, good for vibration to
7.5g. Of course the instruction set is known, and we know that the logic
portion is largely made from DTL integrated flat-packs, but we have no
circuit diagrams. The sales literature talks of “easily swapped modules”
which is fine if you know which one to swap and have some spares. The
modules are wire-wrapped in, so swapping them to locate a fault is not
really on, and you cannot tell much about what is in them without cutting
them open beyond repair. So it was clear from the onset that getting a
920M working might be a challenge.
Well after my time at Rochester, at least two more versions of the 920M
were produced as late as the 1980s under the Marconi-Elliott or GEC
badges (for export to overseas air forces), namely the 920M MCM7 and
the 920ME, neither of which I knew anything about until a couple of years
ago. The MCM7 uses the same construction as the MCM2 & MCM5, “built in
three hinged sections for maximum servicing accessibility” opening up to
form a Z or N. The 920ME is an almost contemporary plug-in replacement
for it, but of completely different construction with space to spare. Gone
are the Araldite modules, there is a small rack with just a few slot-in easily
removed cards using AMD 2900 bit-sliced chips. Gone too is the core store,
to be replaced by a CMOS memory and battery backup — remember that
the flight program had to be retained between missions, and was only
8 Resurrection 90 Summer 2020
reloaded by wheeling a PLU (program loading unit) to the aircraft when
the software was updated.
So what has been happening? Those of you who were on the CCS trip to
Munich in April 2017 will remember meeting Dr Erik Baigar, who is
something of an Elliott enthusiast. He has owned two 12-bit Elliott airborne
computers since 2004, and in 2016 he bought a 920ME from eBay (the
first time I’d ever heard of them, see Resurrection 80) and, after a little
trouble with the power supply and memory controller chips, he got it
working. So when I visited him in 2017 I was able to run some test
programs on it.
Then, shortly after we received the RAA’s 920M MCM7, another 920ME and
three 920M MCM7s turned up on eBay, which Erik purchased, sending
MCM7 number 5343 marked JAGEX to me. Erik was able to get his
(second) 920ME working without too much trouble, given that he already
had a test rig. Getting the MCM7s working has proved to be more difficult,
as we expected. Erik has worked hard on this since the start of 2020,
whilst I’ve provided some support by “ringing through” 5343 (unpowered).
Problems encountered include:
• The covers which hold the three hinged section closed together are
secured by substantial bolts, which came out easily, and by much
smaller countersunk screws, a couple of which had to be drilled out.
• The pin-outs of the 920ME and the 920M MCM7 were essentially the
same, but the power requirements of the MCM7 with its core store are
more complicated.
• The connectors are miniature Deutch DSM-series costing around £1000
each, and each 920M needs at least two. So Erik just bought the right
pins, and he and a friend milled some cylindrical inserts to hold the
pins together in the right place without shorting.
• Neither of Erik’s MCM7s worked when powered up, but instead they
took excess current. The main problem was some store modules
contained banks of hidden small capacitors across the power rails,
which were failing. By some very clever detective work he was able to
see where the current was going, and by cannibalising one MCM7 he
got the other working.
The standard XSTORE store tests, TicTacToe, BASIC , Countdown, and 8
Queens have all now run. So exceptionally well done to Erik! There is
much more, and numerous photographs (far too many to include here) on
Erik’s site at www.baigar.de/TornadoComputerUnit/TimeLine.html.
Resurrection 90 Summer 2020 9
My photograph here
shows 5343, with
covers removed and
sections slightly
opened up. The core
store is in the left
section, and power is
being provided via one
of Erik’s milled plug
inserts. The
FlexoWriter to the
right was one of those
used to prepare the
Jaguar Flight Program in around 1970.
I was reluctant to power up 5343, knowing that neither of Erik’s MCM7s
worked initially, and I only have limited test gear (an analogue multi-
meter and a TTL logic probe, no oscilloscope), but 5343 did have a
“serviceable 2/6/87” tag on it. So over Easter I assembled a test rig,
getting -5.5v & lots of +5.5v (for the logic) from my own 903’s power
supply, and getting +12v & +18v (for the store) from a laptop brick, all
wired through the plug insert shown. Next I wired the control signals to a
ThinkPad printer port, also simulating tape reader input and two bits of
tape punch output. (A printer port is designed for 8-bit output, but not for
8-bit input unless it is attached to tri-state logic).
I took the precaution of disconnecting those troublesome capacitors, and
fitting some externally. Much to my amazement and luck, 5343 did power
up. Even more amazing was that it contained a program, in fact, an issue
of the Jaguar Flight Program. And it didn’t take long to find the “present
position” and “first waypoint” in it, both at 17.677°N, 54.028°E. Look this
up if you want to know where this computer was last used. But it is rather
odd that this once-classified code was not wiped. (I arranged for an issue
of the JFP to be de-classified back in 2016, see Resurrection 72).
Excitement was somewhat tempered by the discovery that ⅛th of the store
was zero (specifically the words 64 to 127 of every 512 words) but it
should be possible to fix this using just one store module from the MCM7
which Erik is now using for spares. Possibly more worrying is that 5343
only runs for long enough to upload a handful of words before I have to
recycle the power: I may need to investigate those capacitors.
So where are we with the TNMoC’s 920M MCM7 number 88, which Andrew
Herbert is currently looking after? Well, just as soon as lockdown is over,
10 Resurrection 90 Summer 2020
we should be able to test it with my breadboard power rig and ThinkPad,
and then we can decide what to do next. We do now understand how most
of the store modules function. The logic sections (especially how the
microcode is implemented) remain largely a mystery, but luckily they
seem to be more reliable. Finally if anybody out there does have the circuit
diagrams, please do let us know!
Software — David Holdsworth
Leo III
I’ve checked the Leo Society’s LEOPEDIA and the CCS links are now
correct.
Ken Kemp and Ray Smith are working on sorting out more demonstration
material for Leo III emulation.
I am encouraged by reports that Delwyn’s Data Recovery activities are
close to reading Leo III tapes. The big hope here is that we shall find the
CLEO compiler.
The latest emulator for Leo III, which introduces separate Java
implementations of the Leo III console and the Leo III printer, is still not
available for Windows. There is low-level activity towards getting a clean
resolution of of the incompatibility between Microsoft’s implementation of
TCP/IP sockets and everybody else’s.
KDF9
Although Bill Findlay and I have been making progress with KDF9 directors,
there is as yet no new end-product in that area. Bill’s version 4 of ee9 has
been compiled for Intel Linux in addition to Bill’s Mac version, so it can
now be run under the Windows Subsystem for Linux (WSL)
Experiments using tesseract 4.0 (the GNU OCR program) on scans of KDF9
documentation have started to yield fruit. The current state of play can be
seen at sw.ccs.bcs.org/KDF9/SRLM-ocr/.
Although tesseract does not retain the spacing within a line, I have been
quite successful in experimenting with code to restore the left-hand
margin. There has been very little manual editing in producing the above
documents. There will be more of them in the future.
Atlas I
The Atlas I emulation project has been dormant for some time. However,
the present situation has afforded Dik Leatherdale with the opportunity to
make some progress.
The emulator “normally” performs input/output to from and to files in the
(Windows) host machine but an inauthentic facility was added many years
Resurrection 90 Summer 2020 11
ago to allow simulated card punch output to be displayed on a desktop
window while the program is running. To this has now been added the
contrary facility to accept data interactively as from a simulated card
reader.
Although a working version of the Brooker-Morris Compiler Compiler was
implemented some years ago, there has been no attempt to make use of it.
But now a small demonstration programme has been successfully created.
The program syntactically analyses a conventional (integer) mathematical
formula and calculates the result. All in less than 120 lines of code
confirming the genius of the original concept.
ICT/ICL 1900 - Delwyn Holroyd, Brian Spoor, Bill Gallagher
ICT 1830 General Purpose VDU
Continuing the work since receipt of the copy of TP4094: Visual Display
(courtesy of the Centre for Computing History, Cambridge) has resulted in
an almost completed change to drawing characters using what is likely a
close analogue of the original method.
In addition, all of the test/demonstration programs we have now appear to
run satisfactorily under GEORGE 3 running on the 1904S emulator.
Work is still on-going to reverse engineer the programs in order to
recreate the original subroutine libraries as far as it is possible.
ICT 1905 Emulator
The 1830 has been added to the 1905 emulator, the only known
installation of an 1830 was with a 1909 in Berlin.
Most problems with the 1974 magnetic tape emulation are now fixed,
certainly under E4BM, further E6RM testing is needed.
In addition a few minor bug fixes/enhancements made.
E4BM Executive for 1905
A new executive package to support the 1830 is currently being written,
the original package is lost to the mists of time. This package is being
based on the existing 1938 Interrogating Typewriter package as the
requirements/functionality are similar.
Currently working modes are: open, close and write to 1830. The read and
combined write/read modes need to be completed for full functionality.
We can now successfully generate a new E4BM under E4BM.
12 Resurrection 90 Summer 2020
ICL 1904S Emulator
The 7930 scanner emulation is now able to write data to its lines. Reading
data is not yet working. A lot more work and testing is required.
In addition a few minor bug fixes/enhancements made.
Batch Application Demonstration System
With the application of WD40 to Brian’s COBOL memories further progress
has been made on an Order Processing system (mark 1) for a fictitious
mail order company. The purpose of this system is to show what was
required to process data back in the days when your input medium was
cards, master files held on magnetic tape and everything processed in
batch runs.
This is very much an on-going longer term project as a reminder of how
things used to be, before the days of databases and online access. Once
the current magnetic tape mark 1 system is completed, a mark 2 is
envisaged moving the master files to disc, then a possible mark 2A version
introducing VDUs and online order entry.
ICL 1900 Website (www.icl1900.co.uk)
More manuals added to the Programming and Applications sections.
ICL PF50
Further to the article by John Harper in Resurrection 89, Brian wishes to
make it clear that he covered the cost of shipping the E4BM
documentation to Bill for scanning — many thanks to Jacqui Garrad at
TNMoC for her assistance in this matter.
Harwell Dekatron/WITCH — Delwyn Holroyd
The machine has been running well since the replacement of the HT power
supply choke. A number of failed store anode resistors have needed
replacement plus a coupling diode. The start switch on the control panel
has also been replaced.
CCS Website Information
The Society has its own website, which is located at
www.computerconservationsociety.org. It contains news items, details of
forthcoming events and also electronic copies of all past issues of
Resurrection, in both HTML and PDF formats, which can be downloaded for
printing.
At www.computerconservationsociety.org/software/software-index.htm
can be found emulators for historic machines together with associated
software and related documents all of which may be downloaded.
Resurrection 90 Summer 2020 13
News Round-Up
Tim Denvir writes to point out that this year is the 60th birthday of Algol 60.
He opines that “Algol 60 at 60 is a pleasing assonance”. We covered Algol
60 in some detail on the occasion of its 50th in Resurrection 50 — more
assonance — so this year we just note it in passing with a smile.
101010101
Guy Haworth is researching the AEG-Telefunken TR4 computer of the
1960s. If readers have any documentation beyond that held at Bitsavers
then Guy would be grateful. Contact via the editor please.
101010101
There have been a number of passings of notable IT people. Margaret Sale
is remembered on page 25, but sadly there are more —
• Professor David Aspinall, best known for his contribution to
Manchester University’s Atlas computer, passed away in December.
• LEO stalwart Tony Morgan went in March. Tony it was who had the
idea of DME/LEO, a microcode for the ICL 2960 specifically to
extend the life of the Post Office telephone billing system which
previously ran on LEO 326s.
• Jack Schofield, the Guardian’s former computer editor and author of
its unmissable technology advice column, Ask Jack, in April.
• John Conway, distinguished mathematician, but chiefly famous for
his Game of Life.
• Finally, Olaf Chedzoy who presented his experience of programming
the Ferranti Mark I to the London CCS meeting in January sadly
died only two weeks later.
101010101
Brian Aldous writes to tell us that the treasure trove which is
Communications of the ACM is now freely available at dl.acm.org/loi/cacm.
Stretching back to January 1958 and forward to the current issue it forms
an unparalleled record of the development of our discipline.
Resurrection Subscriptions
This edition of Resurrection is the last in the current subscription period.
Go to www.computerconservationsociety.org/resurrection.htm to renew
for issues 91-94, still at the modest price of £10. New subscribers are, of
course, welcomed.
14 Resurrection 90 Summer 2020
Packet Switching and the NPL Network
Peter Wilkinson
At the back end of 1966, Donald Davies, then Superintendent of the
Computer Science Division of the National Physical Laboratory, was
authorised by NPL management to set up a small team to
investigate the feasibility of packet switching (see later) as a means
of interconnecting computer services. Davies had first proposed the
basic idea in a short memorandum published in 1965.
Originally, Davies had been recruited to NPL in around 1947 to act as an
assistant to Alan Turing, who was to join the laboratory from Bletchley
Park in order to design and develop an early computer. Turing quickly
became disillusioned with the civil service and departed for academia,
leaving Davies as a key member of a team aiming to realise Turing’s
outline design ideas, first in the form of Pilot ACE, which later transformed
into the full ACE (and also spawned an early commercial machine, DEUCE).
ACE was still providing a computing service to the NPL in 1966 although by
then it had been joined by the English Electric KDF9.
By the mid-1960s the first generation of computers was being replaced by
more powerful ones which did not require a team of hardware engineers to
keep them running 24/7 (remember “powerful” is a relative term, probably
still less than a modern wrist watch!). New techniques such as virtual
memory and time-slicing were coming into play, allowing machines to run
several jobs (programs) simultaneously. In the USA around this time,
Project MAC was being developed at MIT to allow computer users direct
access from terminal devices (teleprinters or even display/keyboard
combinations) to load and run programs under their own control and
receive the resulting output directly rather than via some intermediary
operator. Some of these terminals were to be remotely sited and
connected via dedicated landlines hired from one of the telephone service
providers. The nascent but growing computer services industry was also
interested in such communications services so that, for example, rather
than collecting application data at various local sites, storing it on
magnetic tape and then transporting said tapes by road to a processing
centre, the data could be transferred directly over dedicated
communication links to a processing centre.
Such communications facilities were being provided by local
telecommunications operators whose facilities were primarily aimed at
speech (telephony) traffic. A dedicated ’phone line would have a modem at
either end to match the digital traffic to a speech circuit and would, at that
Resurrection 90 Summer 2020 15
time, typically run at 600 or 1200 bits/second, far too slow for the transfer
of large files of data or to enable high speed interactions.
Davies recognised that speech networks, apart from having too little data
bandwidth, were badly matched to the needs of such bursty data traffic
and that the evolving situation called for a different type of network that
could allow long messages (e.g. files of data) to be transferred between
multiple sources and destinations while at the same time allowing fast
responses for real-time interactive traffic. A speech network dedicates a
circuit for the duration of a call whether or not there is meaningful traffic
on that circuit, while setting up and clearing down each circuit then took
several seconds, far too slow for interactive traffic. In his 1965
memorandum, Davies envisaged a dedicated network of high-speed links
interconnected via switching centres (“nodes”), themselves based on small
computers, that would transfer short message blocks (which he named
“packets”) from node to node across the network. Each packet would
include source and destination addresses, error correcting code and some
limited management information. The nodes would be provided with
routing tables to ensure that each packet was directed along the most
efficient route, with alternatives to cope with failure situations. The packet
switched network can therefore be regarded as a mechanism for
transferring streams of interleaved packets simultaneously between
multiple sources and destinations; because packets are short and link
speeds high, both good transfer rates and short response times are
enabled.
Originating and receiving computers, themselves connected directly to the
packet switches, would include extra management information in the
packet data field to allow them to, for example, collect a number of
packets together into one longer message by means of a message
identifier and a sequence number. The packet switching network would be
entirely transparent to such higher-level protocols.
Another very significant advantage of the packet switching technique is its
flexibility. As packets are held in the memory of each node computer
before being forwarded to the next, it becomes relatively simple to
upgrade the network by increasing link speeds in a piecemeal fashion.
Adding extra links, nodes and user computers mainly becomes a matter of
updating the routing tables. This flexibility has been evident in the modern
internet, which has grown enormously in terms of speed and capacity yet
in a manner almost invisible to its users.
The team Donald Davies assembled in 1966, under the management of
Derek Barber, consisted of Roger Scantlebury, Keith Bartlett and Peter
16 Resurrection 90 Summer 2020
Wilkinson. Its remit was to undertake a feasibility study of a possible UK
national-scale packet switched network comprising 12 nodes covering 12
major centres (all of which had universities with significant computing
facilities). The inter-node links were assumed to run at 1.5 Mbits/sec
based on a planned trunk network upgrade by the Post Office. The Post
Office at that time was a department of government providing both the
mail/telegram and speech telephony services.
The feasibility study made an outline design of the link protocol and
routing algorithm, considering three possible configurations for the node
computer, with maximum, intermediate and minimal link hardware to
assist the node computer to run the protocol, input and error-check the
packets. At the same time the node computer software processing the
packets and handling their routing was sketched and its performance
estimated for a then typical process-control minicomputer. The
intermediate configuration proved most cost effective, with the node
computer capable of handling around 250 packets/sec.
Using this data, a simulation of the complete 12-node network was
developed (by Roger Healey) and its performance estimated for various
levels of offered traffic. This gave some idea of the overall throughput and
response time achievable using the full network, which were generally
acceptable for moderate traffic levels. However, as offered traffic ramped-
up, the worrying discovery was made that the network became saturated
and throughput collapsed. Although a shock at the time, with the benefit
of hindsight it is obvious that something like that must happen. The
simulated network is overconnected with several possible paths between
most source/destination pairs (essential if the system is to be resilient to
link or node failures). Normally a packet held in any node is forwarded via
the link giving the most direct route to its destination. However, when a
long output queue is detected, the node forwards via an alternative link
with a shorter queue on the assumption that it will thus reach its
destination faster. This “myopic” alternative routing mechanism was
adopted for simplicity since it does not require the network to develop an
overall picture of system status. The consequence is that, just as the
network is becoming busy, packets are being forced to take longer routes
hence the amount of work needed to process the same level of traffic
begins to increase.
Donald Davies was concerned by this finding and, typically inventive,
proposed a solution (which he termed “isarithmic”) to keep offered traffic
levels within bounds, based on the use of permits (or tokens). No sender
could introduce a packet to the network unless it possessed a token. Once
a packet had reached its destination the associated token would be re-
Resurrection 90 Summer 2020 17
used locally or redistributed according to some algorithm. The number of
tokens available in the network was limited in order to maintain moderate
traffic levels. Obviously the isarithmic mechanism would need careful
management in order to avoid overly restricting system throughput and in
itself represents an additional system overhead. The issue of network
traffic load management is a significant one which has a range of possible
solutions. Donald subsequently pursued these questions by setting up a
specific network simulation group under Wyn Price.
The NPL’s more detailed network proposals were described in a joint paper
which was presented by Roger Scantlebury at a conference in Gatlinburg in
1967, where a paper summarising the plans for a broadly similar US
network, which became known as the ARPANET, was also presented by
Lawrence Roberts of the Department of Defense, but without defining its
intended communications technology. The upshot, resulting from
discussions at that conference, was that packet switching was adopted for
the ARPANET. The ARPANET of course became the principal forerunner of
the Internet.
In the USA there appeared no significant barriers, either financial,
organisational or in motivation or skills, to taking these ideas rapidly
forward and DoD contracted the engineering firm Bolt, Beranek and
Newman to build the ARPANET packet switched communications network.
In the UK, by contrast, such barriers proved insurmountable, at least in
the short/medium term.
Over the period 1965 to 1968 Davies had liaised closely with the UK IT
industry and with major players such as senior staff in Post Office
(Telephones), holding several conferences and meetings on the subject of
data networking. Many individuals within the IT industry were strongly
convinced of the need for a national network and indeed formed a lobby
group (the Real Time Club) to promote such ideas. However, while
showing definite interest in the technical concepts (to the extent that they
later agreed to second one of their engineers, Alan Gardner, to the NPL
team) the GPO was unconvinced of its economic viability, believing that
speech traffic would always dominate data traffic. It was therefore
reluctant to sanction use of its trunk network technology as the backbone
for a national data network. The Ministry of Technology (NPL’s then
overlord) was in no position to insist and passed the buck to NPL.
That left only the possibility that the NPL team could demonstrate the
feasibility of packet-switching by building a purely local network to cover
the NPL’s own campus (then about 78 acres and half a mile in width). This
newly proposed NPL network was to provide a range of computing services
18 Resurrection 90 Summer 2020
(such a storage for files) as well as basic communication facilities, so that
many types of peripheral devices and existing computer services could be
interconnected. Costs were to be covered from within NPL’s existing
budgets and management sanction was obtained.
In Davies’ early network proposals, connection of devices and services
locally to each other and to the high-level packet network was to be made
by an interface computer. Thus the task now was to fully define the
functionality of an interface computer for the NPL and start its
implementation, which began in earnest in mid-1968. The NPL campus
was already provided with a backbone network of co-axial cables capable
of rates of 1 Mbit/second, then incomparably faster than speech network
modems. Consideration was being given to adopting a newly-designed
British minicomputer, the Plessey XL12. The XL12 had an ideal
input/output mechanism capable of attaching up to 512 terminal devices
via a “demand-sorting” system which transferred data from attached
devices directly into computer memory. Alas, a government-inspired
reorganisation of the UK computer industry prompted Plessey to cancel
further development of the XL12 in the summer of 1968.
The NPL team had earlier undertaken an evaluation of available
minicomputers in arriving at its decision to buy the XL12. The Honeywell
DDP516 was chosen as next best, in part because it also had flexible
direct-to-memory input/output facilities. This turned out to be a good
decision, not least because the Honeywell engineers were interested in the
project and willing to adapt its I/O mechanisms to NPL’s requirements.
Incidentally, the DDP516 had also been chosen by the ARPANET team as
its packet-switching node at around the same time and for similar reasons.
The first DDP516 arrived on site in early 1969 (another was later
purchased as a backup).
At a meeting in Barber’s office in the autumn of 1968, early progress on
the network development was discussed. Scantlebury and Bartlett had
already been designing some aspects of the communications hardware and
Wilkinson was formally tasked with producing the corresponding software
for the DDP516. However, at that stage there was no detailed specification
of the network’s overall functionality and major aspects were still up in the
air.
Some decisions had already been taken regarding the transmission
hardware, including the crucial one that the network should offer an
interface to its subscribers based upon British Standard BS4421, in whose
creation Derek Barber had played a major role. Its key features are that it
is a fast parallel handshaked interface, the latter being essential for flow-
control. During 1967/68, Scantlebury and Bartlett had designed a link
Resurrection 90 Summer 2020 19
control mechanism for use in the network, allowing reliable full-duplex
transmission between a master/slave pair of “line terminals” at either end
of one of the co-axial cables.
Another task underway was the design of a multiplexing scheme to replace
that lost with the demise of the XL12. The plan was to have 8-way
multiplexers that could be stacked in up to three layers, permitting a
theoretical maximum of 512 connected terminals. Each multiplexer
contained a demand-sorter to arbitrate clashes between its inputs. Each
multiplex layer added a three-bit address to its output to the next higher
level, so the top level multiplexer would present to the computer a data
byte from the originating terminal accompanied by a nine-bit terminal
address. The non-standard Honeywell hardware (developed during 1969)
would then use that address to store the terminal data in an allocated
buffer in the DDP516 memory. Multiplexers could either be interfaced
directly to their predecessor/successor using a parallel BS4421-like
connection, or be joined via a serial link using a line-terminal pair. The
system thereby offered considerable flexibility.
The question still unaddressed was — how should the communication
between subscribing terminals on the NPL network be organised? Packet-
switching is a radical departure from the traditional telephone system
whereby one subscriber makes a call to another, setting up a circuit for
that purpose. Paradoxically perhaps, that was also seen as the natural way
to organise the NPL network; if one terminal wishes to interact and
send/receive data with another then the network should permit them to
establish a virtual call. “Dial”, specify the destination “number”, if available,
establish the “connection”, interchange data (in the form of packets,
because that was how the central computer server was intended to
operate) and finally end the connection when no longer needed.
This approach was thrashed out by all concerned and made apparent the
need for a third type of hardware module, a network termination unit. This
was named the Peripheral Control Unit (PCU). The PCU offered the BS4421
interface to subscriber equipment and would itself interface directly, either
to a multiplexer or to a slave line terminal unit. The PCU was equipped
with a small specialised keypad comprising four buttons with lights, by
means of which the operator at a subscriber terminal could establish and
manage a virtual call to another. For simple devices a numerical keypad
was also provided to allow input of a destination address. It was now
necessary to alter the communication hardware designs so that both user
data bytes and virtual call status information could be transferred to the
central computer. A ninth data/status bit was added, so that in total an
18-bit unit would be received at the centre.
20 Resurrection 90 Summer 2020
With system design complete, it was possible for hardware and software
teams to concentrate on their specific areas. Thus in early 1969, design
turned to implementation. The hardware team under Keith Bartlett’s
leadership undertook final detailed specification, implementation and
testing of master/slave line terminals, multiplexers and PCUs. Bartlett’s
team comprised Les Pink, Patrick Woodroffe, Brian Aldous, Peter Carter,
Peter Neale and a few others. They also liaised with Honeywell in the
adaptation of the DDP516 input/output controller.
On the software side, Peter Wilkinson was assisted by Keith Wilkinson (no
relation) and Rex Haymes. One downside of the choice of the Honeywell
computer was that it came with very little software. There was an
assembler/loader for its simple assembly (low-level) language, DAP16 and
a real-time executive, which a quick assessment suggested was aimed
primarily at process control applications (such as running a chemical plant).
It did not offer an appealing mental model for how the software of the
network computer should be organised. There were no readily available
publications on communications software and the like, as this was such a
new field, so Wilkinson was faced with a blank sheet of paper — actually a
good place to start as there are no constraints!
It was essential to find some general organising principles that would help
to structure the task. In that, Wilkinson was fortunate. Before joining the
network group he had worked in the programming research group, with
Mike Woodger as a colleague. Woodger had joined NPL at about the same
time as Davies, also to be an assistant to Turing. He had also worked in
the ACE computer team and later became one of the authors of the high-
level programming language Algol 60. He had subsequently continued as a
member of a committee defining an even more powerful language, Algol
68, where one of his colleagues was the computer scientist Edsger W
Dijkstra, then working at the Technical High School in Enschede. Dijkstra
was a proponent of the notions of process and of hierarchy and had used
these in the development of a model operating system, of which he had
produced an elegant description, available through Woodger.
A process is an abstraction, ideal in situations where a single central
computer is dealing with an external environment in which many events
can arise independently. External events can be represented by distinct
processes in the computer; they share its central processor and other
resources and communicate with each other in order to carry out the
overall computing task required. All that is needed is a simple executive
which ensures that each process is given a slice of computing and other
resources. Hierarchically structured software is organised into distinct and
Resurrection 90 Summer 2020 21
well-defined layers in which each layer provides services to the layer
above via a clearly defined interface.
Wilkinson decided to structure the network computer software as a small
number of processes, each carrying out particular tasks. It was further
decided that these processes would communicate via an internal message-
passing mechanism (by analogy with the network itself). An executive was
designed with layers for process scheduling, message queue handling and
memory management (dealing with packet buffer allocation to receive and
forward subscriber traffic). The DDP516 generated an interrupt each time
a status byte was received, which the executive software turned into a
new message, queued for the appropriate process.
The software for the various layers and processes was written using the
high-level language PASCAL (related to Algol 60) and then hand-translated
into DAP16 by the team. The software produced by this approach was
highly modular, so that useful testing could be carried out on each unit in
isolation. In this way, when the full system was assembled, most of the
errors had been eliminated and the final integration testing was relatively
painless.
The network software came into operation in January 1970 and an
informal network service was initiated, though at that time few if any of
the services intended to be accessed via the network were available.
However, once the system had reached an acceptable level of reliability,
some performance measurement could be undertaken. The outcome was
rather disappointing, as the switching computer was only capable of a
maximum throughput of around 150 packets/sec. This was mainly a
consequence of the high processing overheads introduced by the message-
passing mechanism. Nonetheless the performance was adequate for the
initial trial uses of the system.
During 1970 various services and facilities were added to the network to
make it attractive to its users throughout the Laboratory. The KDF9
computer service was connected, allowing remote access and job entry; a
contractor was engaged to produce the software for a filestore service on
another DDP516; a new information retrieval system, being developed by
another group in Computer Science Division under David Yates, was
linked-up to provide access to its users.
While the NPL network was beginning to prove its worth, besides low
packet throughput, another limitation was becoming more apparent: all
subscribers, including computer services, were connected as simple
terminals. If a computer service was to deal with several simultaneous
users, it therefore required several network connections via PCUs, an
22 Resurrection 90 Summer 2020
inefficient use of hardware and an inflexible approach overall. It would be
far preferable to allow the service computer to handle the virtual call
mechanism by software and to provide it with a link to the network at the
packet-level. These limitations led to a rethink and a plan to upgrade the
network service from what in 1970 was described as the “Mark I” to a
shiny new “Mark II”. Mark II came into full service in 1973 but that is
another story, as they say.
At a distance in time of over 50 years, memories fade, become unreliable
and detail starts to blur. Needless to say, this account is not just made
from memory. The NPL team made many publications along the way. Most
particularly, in 1985/86 they invited Martin Campbell-Kelly then of
Warwick University to meet with them, review all their activities, conduct
interviews and produce a report overviewing the work of the
communication network team over the period 1965 — 1975. That report
has guided this article for Resurrection. [Data Communications at the
National Physical Laboratory (1965–1975). Annals of the History of
Computing , 9 (3/4), 1988]. That report contains many of the key
references, for anyone wishing to follow up the details.
On a personal note I look back on the NPL network project as the most
exciting and fulfilling of my life, working with a team of excellent
colleagues and doing something which was both fun and challenging, with
a lasting impact. Of course, it is fair to say that no one, not even Donald
Davies, had even the haziest idea of what the Internet would eventually
become. It is also paradoxical to realise that while speech traffic certainly
dominated the communications scene in the early 1960s, the boot is now
completely on the other foot, as speech networks are now packet-based. I
also think it is true that the Internet, and particularly packet-switching
owes more to Donald Davies than it does to any other single person.
In summary, the NPL campus network (Mark I) was the first digital local
area network in the world to use packet switching and high-speed links.
Resurrection 90 Summer 2020 23
Thoughts on the ICL Basic Language Machine
Virgillio Pasquali
In Resurrection 87 we published an account of the deliberations
which led to the ICL New Range of the 1970s. Reader Mike Hore
wrote to the author to discuss the reasons why John Illiffe’s Basic
Language Machine was not selected as the basis for New Range.
The response is interesting and so is reproduced here. [ed]
I cannot help feeling that you are asking the wrong person why ICL
selected the Synthetic Option instead of the BLM. The Jury, in October
1969, spent more than a week secluded in Cookham conference centre,
investigating the two proposals and interrogating people. The top 12
technical managers in ICL, under the chairmanship of Colin Haley, were
certainly capable of forming their own opinion based on what the company
needed at the time. I did not participate in their deliberations, for obvious
reasons, and the Jury never published their detailed analysis or discussed
their findings with me (I was not part of the Jury as I, like Roy Mitchell of
the BLM option, was the leader of one of the two options on trial).
But I remember that the choice was not predetermined. They looked
carefully at both the options and arrived at their own conclusions, and
clear recommendations.
But I will tell you why I thought, and I still think, that they made the right
choice. It is my own view, and I do not know if this is the reason why ICL
did not adopt BLM.
It can be summarised by relating a conversation that Mike Forrest and
Gordon Scarrott had while we were having a drink at our local pub after
work. Gordon said to Mike “Why are you building a road while you have a
motorway available to you?” and Mike replied “If the motorway does not
take you where you want to go, it is of little use”.
BLM, conceived by John Iliffe while at Rice University in the USA, was a
great technical idea that did not fit the market. (Many years later I had a
similar experience with CAFS. A brilliant idea, and I insisted it should be
marketed against the opposition of the Data Management team. It was not
a success in the market, and we lost money and, more importantly, we
wasted scarce resources. But even today we technical people talk of CAFS
as a great product).
I do not remember the technical details, but a few relevant facts are:
1. ICL had about 2% of the world market, it was starved of capital, its
home market (the UK market) was far too small, and access to the rest
24 Resurrection 90 Summer 2020
of the market very poor (David Talbot expressed it very clearly in his
article in Resurrection 35). On top of that, our manufacturing costs
were high , resulting in poor profitability and low margins for the lower
part of our mainframes, making it highly desirable to have a New
Range capable to excel at the top end of the range (Hence the top
down approach and the emphasis on the MU5 architecture as the
starting point in the evolution of the Synthetic Option architecture).
The 1900 architecture had lost ICT most of the highly profitable large
machine market and we wanted to regain it with New Range.
2. I am not a marketing expert, but I learnt very early on that a market,
once established, will not easily accept discontinuities that endanger
(large) existing investments by the supplier and, more importantly, by
the customers. (The Japanese 5th generation programme confirmed
this as did our experience with CAFS). The user wants to be safe. The
360/370 architecture was rather pedestrian, but it was solid and safe
and it went on to conquer the world. The IBM technical people tried to
introduce a new architecture (FS) at approximately the same time as
ICL and their marketing people vetoed it. But we were forced into a
new range by the merger, when we found out that the System 4
strategy was not feasible.
3. The new range would have to co-exist with the mainframes of other
competitors (IBM etc.) possibly on the same site, and the ability to
exchange data easily with other alien mainframes was important. It
was highly desirable to have common international standard data.
Tagging the data inside the data format was undesirable (the Synthetic
Option wanted to absorb the powerful and secure addressing structure
of BML, but we kept the data formats to international standards and
put the tagging in the descriptors, and then we had to weaken it to
allow alien programmes to run in their Virtual Machine on top of the
Kernel, if I remember correctly).
A number of articles in Resurrection 35, attempting to analyse why the UK
computer industry did not succeed (in spite of the creativity of our
technical people and excellent technologies), could be helpful to you in
giving an overall perspective of the issues.
One final point: when ICL failed to adopt the BLM in 1969, I told John Iliffe
that he could seek another company to adopt BML. ICL would not impose
any property rights and we would give it away for nothing.
But, as far as I know, John could not find anybody interested, not even in
the USA where there was plenty of capital to exploit innovation (while we
were starved of capital and resources).
Resurrection 90 Summer 2020 25
Obituary : Margaret Sale
Stephen Fleming
Born in 1932, Margaret trained as a
school teacher in the post-war years
and brought up a family with her
husband, the late Tony Sale (who later
led the rebuilding of Colossus the
wartime codebreaking machine).
Margaret’s interest in Bletchley Park
and cryptography was awakened by
talking to guests at a 1992 gathering of
Bletchley Park veterans when
redevelopment of the Bletchley Park
site seemed imminent. She then
became a founding member of the
Saving Bletchley Park campaign and
joined the board of Bletchley Park Trust
in 1994, serving for six years. In
addition to leading some of the tours,
she undertook fundraising, set up
lecture series and conferences.
Volunteering her time and boundless
energy free of charge, she held important roles on the Park for three
decades, being present on site most days until quite recently. Speaking
shortly after the death of her husband Tony in 2011, Margaret told a radio
interviewer, “Tony and I were known as Mr and Mrs Bletchley Park.”
When Tony Sale and his team began the rebuilding of Colossus to honour
the work of the wartime men and women on Bletchley Park, Margaret
worked diligently in the background facilitating the project. She was
thrilled when the new Colossus Gallery was opened at The National
Museum of Computing in 2012.
A supporter of TNMOC since its inception, Margaret joined its trust board
in 2011 and up until a few months before her death was to be found most
weekdays engaging visitors in the Colossus and Tunny galleries at the
Museum.
In 2017, she received a Point of Light award from the prime minister in
recognition of her outstanding contributions as a volunteer and in 2012
she received a Sunday Telegraph/Waitrose British Volunteer Award.
26 Resurrection 90 Summer 2020
Why is there no well-known Swiss IT industry?
Herbert Bruderer
Computer science in Switzerland began in 1948 with the founding
of the Institute for Applied Mathematics at ETH Zurich. Eduard
Stiefel was the director, the two assistants were Heinz Rutishauser
(chief mathematician) and Ambros Speiser (chief engineer).
In 1950, ETH Zurich was the first and only university in continental Europe
with a functioning punched-tape controlled computer, the Zuse Z4 (built in
1945, restored and expanded in 1949). In Sweden, the plugboard-
controlled Bark relay computer was put into operation in 1950. The Swiss
Polytechnic used the leased mechanical relay computer from 1950 to 1955.
In 1951 the trio published a fundamental work on computer construction.
One year later Rutishauser’s postdoctoral thesis on automatic
programming (program production with the help of the computer) was
published. He was one of the fathers of the Algol programming language.
In 1956, the self-built Ermeth electronic vacuum tube computer was
operational. Despite these good conditions, no IT industry emerged in
Switzerland in the 1950s. This has often puzzled scholars. What were the
reasons?
Bern Hasler AG wanted to market Ermeth worldwide
The company Hasler AG, Berne (today Ascom), was involved in the
construction of the electronic computer and also provided financial support.
It has only been known for a few years that Hasler wanted to market the
expensive machine worldwide. To this end, it concluded a licence
agreement with the ETH in 1954 for the magnetic storage drum, which it
had not helped to construct. The development of the drum memory device
from left to right Eduard Stiefel (head of the institute), Heinz Rutishauser (chief mathematician), and Ambros Speiser (chief engineer)
Resurrection 90 Summer 2020 27
had caused great difficulties. However, Speiser left the ETH at the end of
1955 — before the completion of Ermeth. Surprisingly, he became the
director of the newly founded IBM Research organisation near Zurich. This
thwarted Hasler’s plans. At this time, two other US research laboratories
were coming to Switzerland in addition to the blue giant: Radio
Corporation of America (Zurich) and Battelle (Geneva). IBM recruited
specialists with high salaries. Speiser’s departure for the competition led to
a discord between Hasler and ETH. On the instructions of the President of
the Swiss School Board, Speiser had to make an appointment with the
General Director of the Berne company. Hasler lost interest not only in the
reproduction of the magnetic storage drum, but also of the entire
computer. This background has only recently become known thanks to
finds in the ETH university archives. A detailed account of the dispute
between Speiser, Hasler and ETH can be found in Milestones in Analog and
Digital Computing, volume 2.
Zuse Z4 relay computer (1950)
28 Resurrection 90 Summer 2020
Both machines have been preserved. The Z4 is now located in the
Deutsches Museum in Munich, the Ermeth and its drum memory are on
view in the Museum für Kommunikation in Berne.
Later attempts also failed
Even later it was not possible to build Swiss computers in large numbers.
This is true e.g. of the Cora transistor computer (1963) made by
Contraves, the Lilith workstation (1982) of Niklaus Wirth and the
Gigabooster super computer (1994) by Anton Gunzinger.
Today, in addition to Logitech, there are numerous research centres in
Switzerland run by Internet giants such as Apple, Facebook, Google, and
Microsoft.
This article appears by permission of Communications of the ACM and the
BLOG@ACM (see cacm.acm.org/blogs/blog-cacm/242462-why-is-there-
no-well-known-swiss-it-industry/fulltext. Photos (except that of the drum)
are courtesy of ETH Library, Zurich, image archive.
Herbert Bruderer is a retired lecturer in computer science an ETH Zurich
(Swiss Federal Institute of Technology) and a historian of technology. He
was co-organiser of the International Turing Conference at ETH Zurich in
1912. Email: [email protected].
Ermeth vacuum tube computer (1956) Magnetic drum memory of the Ermeth (1957) (credit: Radomir Novotny, electrosuisse)
Resurrection 90 Summer 2020 29
Ho Hum, Nothing to Do
Michael Clarke
This is perhaps just a whimsical reminiscence of days gone bye, but one
never knows. Let me start with what prompted this look at computers with
nothing to do. I am a Beta tester for a commercial gaming simulator
product, I have a high-end Intel i7 PC with lots of RAM and a good
graphics card, and a venerable yet strangely still competitive Intel
Q6600PC (2008 vintage) with slower and less RAM and an older, slower
graphics card. The old machine is my email and news sniffer machine
always looking at the WWW. The i7 PC is often just doing testing and
development of scenarios that I create for the simulator. From time to
time I use the simulator’s export/import process to transfer my test
scenarios to the Q6600 PC and check that they run correctly. The old
machine is very much at the bottom range of recommended systems
needed to run the simulator and should be using lower display levels,
shadows, viewing distance etc. I have found that the old machine works
with no discernible difference even when set to the highest display levels.
So, I investigated, and the result was what led me to the title of this item.
‘Ho Hum nothing to do’ is a euphemism for the computer operating
systems scheduler and my first involvement with schedulers at RRE
Malvern way back in 1968. RRE was developing the Algol68R compiler. The
1907F had passed rigorous commissioning tests in September every
peripheral was brand new and customer engineers had very little to do
except scheduled weekly maintenance so, I was learning to program
computers — a few Executive mends, just a few lines of code, hardly stuff
to learn from so I used its Algol compiler to work out my expense sheet for
mileage travelled on company business, which was complicated and
needed to be worked out every week. I refined my program down to a
single statement with lots of declarations, subroutines and a simple data
file about 200 lines of actual Algol instructions and when compiled about
6Kw of object code. It often required changes as I was not that good at
writing programs so it needed to be re-compiled weekly!
This I did after the RRE work had ended or before RRE staff started work,
depending upon which shift I was working that week. I was therefore
running this compilation in a very powerful dual processor (1907F)
computer that was doing nothing else. Well most of the time; occasionally
after a consult with the RRE people as to why my program failed to
compile, or gave rubbish answers, for which they were extremely
appreciative, may I add as I was throwing up all sorts of usage problems
with their compiler. I noticed that the total elapsed time to compile my
30 Resurrection 90 Summer 2020
thousand or so lines of Algol instructions was the same when running
alongside the normal RRE work load and in my solitary runs in an empty
machine.
I bought this strange phenomenon to the attention of the RRE people and
the George III support person (David Owen) and we discovered that the
George scheduler which worked essentially on a per minute slot time was
doing exactly that. My compile required just a few seconds of computer
time but, was being given quite small scheduled time slots and between
George III and the Executive which had a 200ms timer interrupt, were
going into this ‘Ho Hum nothing to do’ hiatus, where the Executive knew
that it had no active process but had been told only to wake up George
once a minute when the real time changed!
This strange effect was quickly addressed and normal throughput at RRE
increased slightly and my expenses program compiled in seconds rather
than minutes in an empty machine.
This brings me back full circle to the remarkable performance of my
venerable Intel Q6600 PC. I turned on Windows 10 Pro Task Manager’s
performance monitoring. When running my most demanding scenario with
intensive display and AI loads, on the i7 It was using about 41% CPU the
Q6600 was using about 75% CPU.
I have not investigated further but, it would seem that ‘Ho Hum nothing to
do’ is alive and well in Windows 10 Pro.
Contact Details
Readers wishing to contact the editor may do so by email to
[email protected], or by post to 124 Stanley Road, Teddington, TW11
8TX.
Members who move house or change email address should go to
www.computerconservationsociety.org/membership/membership_general.htm.
Those who are also members of BCS, however, need only notify their
change of address to BCS, separate notification to the CCS being
unnecessary.
Queries about all other CCS matters should be addressed to the Secretary,
Rachel Burnett at [email protected], or by post to 80 Broom Park,
Teddington, TW11 9RR.
Resurrection 90 Summer 2020 31
50 Years Ago …. From the Pages of Computer Weekly
Brian Aldous – TNMoC Archivist
Ferranti win Argus orders from W Germany: Two large export orders for
the supply of Argus 500 computer control systems to Kalle AG, a major
West German chemical manufacturer, have been received by the
automation systems division of Ferranti Ltd. The first order covers
equipment to enhance the direct digital control capability of an existing
Argus 500 system as well as providing for the simultaneous running of off-
line programs. The other order is for a much bigger dual processor Argus
500 direct digital control system. (CW 4/6/70 p2)
ICL install Etom 2000 at Stevenage: Forming part of the on-line phase of
ICL’s project on automated drawing office procedures which is under
development at Stevenage, an Etom 2000 CRT display has been installed,
linked via a 7020 terminal to a 1904E. Graphic Displays Ltd of Luton,
which is now part of Kode International, now has five orders for the low
cost Etom 2000, which ranges in price from about £4,100 for a basic
system to £8,500 for a system which includes features such as a light pen
and vector and character generation. (CW 4/6/70 p3)
CDC goes after Business Market: Ever since the introduction of the CDC
7600 computer it has been Control Data’s avowed intention to extend its
customer base from the scientific research establishments, educational
institutions and public utilities which still account for the bulk of its
installations, to the larger commercial and industrial concerns. Now this
policy has been taken a step further with a series of announcements
designed to demonstrate the flexibility of the computing power which
Control Data can provide, and the ease with which users of 6000 series
computers can switch to the larger 7600 machines. (CW 11/6/70 p1)
A Major Role for Computers on Election Night: Assessing the impact of
politics in computers has become in recent months, with the machinations
of Sub-committee D, almost a full-time occupation. But with June 18th only
a week away, it might be appropriate to take a look at the impact of
computers in politics. Perhaps the area in which the public will be most
conscious of computing in the general election will be on election night
itself. As usual the BBC and ITN networks will be staging an all-night
election circus, each with its own unique selling points to entice the
viewers away from the other. Electronic aids, following recent precedent,
will play a large part in this. Somewhat paradoxically, both networks will
be using the same bureau, Baric Computing Services, as the basis of their
computerised voting analysis and results forecasting operation. Beyond
32 Resurrection 90 Summer 2020
this, however, the two systems will not converge; Baric is supplying two
teams to develop the different systems, and the BBC will use an ICL 1905F
in London and an ICL 4/50 in Manchester, whereas ITN will be using
Baric’s KDF9 at Kidsgrove. (CW 11/6/70 p10)
VDU System deals with orders from Booksellers: Orders sent to Oxford
University Press from booksellers are now being processed on five new ICL
visual display units which have been linked to an ICL 1901A at the OUP
offices in Neasden. With the new system, the customer reference, the OUP
reference and an individual order is typed on to the keyboard. Books are
entered by the first four letters in the author’s name and four letters of the
title. The VDU displays either the only book on the file which corresponds
to the key letters, or a list of all possible titles, so that the operator can
select the correct one and key in the quantity ordered. (CW 11/6/70 p15)
DEC extends range of PDP-11 Peripherals: The range of peripherals for the
PDP-11 computer introduced earlier this year has been extended by Digital
Equipment Co to include a magnetic tape unit, a disc unit and a
lineprinter. The company has also introduced a new disc unit and a
software monitor for its PDP-8 and PDP-12 computers. The tape unit for
the PDP-11 consists of the TC11 control unit and the TU56 dual tape
transport. The TU56 can store up to 262,000 16-bit words on four-inch
reels, and employs a type of redundancy recording in which each bit of
data is stored on two tracks. The TC11 controller permits direct memory
parallel-word access, and provides for bidirectional search, reading and
writing of data. (CW 25/6/70 p25)
IBM introduces System 370 Models 155 and 165 first in powerful new
range: With the announcement of a new range of computers to be known
as the System 370, and the introduction of two central processors in that
range, IBM has at last pricked the bubble of rumour and speculation which
has for many months surrounded any mention of its forthcoming products.
First deliveries of the two new computers, which are known as Models 155
and 165, are scheduled to take place in February and April respectively
next year. System 370 computers will be manufactured in both the US and
Europe. The 155 will be produced at Poughkeepsie, New York, and at
Montpellier in France, while the 165 will be produced at Kingston, New
York, and Havant in the UK. The 165 is, according to IBM, the most
powerful IBM computer ever to be manufactured in Europe and Mr E. R.
Nixon, managing director of IBM UK, said he anticipated it would “make a
very positive contribution to our balance of payments”. The smaller of the
two new computers, the 370/155, is up to four times as fast as the
360/50. The main store, which has a cycle time of two microseconds,
ranges in capacity from 256K bytes to 2,048K bytes — substantially larger
Resurrection 90 Summer 2020 33
than the 512K bytes which is the maximum core size of the current
360/50. The 370/165, which operates at five times the speed of the
360/65, is also fitted with two microsecond core-store, in this case ranging
in size from 5I2K bytes to 3,072K bytes. (CW 2/7/70 p1)
GPO studies plan for Integrated Network: A comprehensive and detailed
proposal for a future integrated data communications system,
incorporating both store and forward techniques and high-speed circuit
switching, has been devised by Standard Telecommunications Laboratories
and British Telecommunications Research in studies commissioned by the
Post Office. The suggested system would be based on 20 to 30 special
computerised data switching exchanges located in major cities. It would be
digital and synchronous throughout, and would utilise the pulse coded
modulation digital transmission technique envisaged for telephony
transmission. Thus it would employ the existing network of lines and
cables. The long-term intention of the Post Office, according to the study
contractors’ proposals, should be full interworking within a digitised
telecommunications network bearing both telephony and non-telephony
data and other services. The proposed network could carry the traffic
resulting from future expansion of the existing telex network, and any
future new telex network. (CW 16/7/70 p1)
Improved PDP-8 model has Faster Memory: The PDP-8/E, a new low-cost
computer which will, in due course, replace both the PDP-8/I and the PDP-
8/L, has been announced by Digital Equipment Company Ltd. Main
features of the new machine are its re-designed architecture and the use
of the MSI (Medium Scale Integration) technology as well as the TTL
integrated circuits which are already used in the PDP-8/I. Some of the
thinking which went into the PDP-11 appears to have influenced the design
of the PDP-8/E: in particular the provision of an internal bus system called
Omnibus, which allows the processor, memory modules and peripherals to
be plugged into virtually any available space in the machine, closely
resembles the Unibus system incorporated in the PDP-11. (CW 16/7/70 p20)
Post Office sets up Trial Switched Network: Two important initial steps
have been taken by the Post Office in its programme to meet the rapidly
growing demand for data transmission facilities. This month the new Post
Office Datel 48K service, the fastest data transmission service in Europe,
which provides two-way transmission at 48 Kbps over rented wideband
private lines, has come into operation. And even more important for the
future, a trial switched network operating at the same speed between
London, Birmingham, and Manchester, has been set up. The Post Office
says that the success of the Datel 48K service will depend largely on the
availability of suitable terminal equipment. The purpose of the trial
34 Resurrection 90 Summer 2020
network, therefore, is to give manufacturers the incentive to develop 48
Kbps terminal equipment compatible with Post Office equipment, and it
provides a manually switched two-way 48 Kbps public transmission system
and an optional simultaneous speech circuit between the cities. (CW 23/7/70
p32)
£7.5m Contract for GEC firm: Five GEC-Elliott Process Automation March
2140 processors, worth about £1,500,000, will be delivered as part of the
£7.5 million contract awarded to GEC Electrical Projects Ltd by the British
Steel Corporation. The contract covers equipment for the BSC’s new three
million-tons-a-year bloom and billet mill at Scunthorpe, which will be the
largest in the UK and one of the largest in the world, forming a key part of
BSC’s £130 million development at Scunthorpe. GEC Electrical Projects will
be responsible for the supply, erection, and commissioning of the electrical
drives and the computer control system for the mill, which is due to be
fully operational by the end of 1972. The five 2140s will be interlinked in a
two-level hierarchy formation in which one co-ordinating processor will link
the production through the entire bloom and billet mill complex, and in
which individual processors will provide control of primary and secondary
bloom mills, a ten-stand continuous billet mill, and two flying shears. (CW
30/7/70 p16)
Siemens system to control London traffic: An important Greater London
Council contract for the automation of traffic control in central London has
been awarded to the German company, Siemens, despite two British
tenders from Plessey and GEC Elliott Traffic Automation. The order, valued
at £759,000, is for two Siemens 306 processors of 32K and 48K 24-bit
word capacity respectively. The configuration will also include two disc
units, a lineprinter, punched card peripherals, and data transmission
equipment which will be used to transmit signals to the 300 sets of traffic
signals. The project will cover an area in the heart of London, between
Victoria (in the south and west), Marylebone Road (north) and Tower
Bridge (east). Data on the occupancy of roads within the network, traffic
flow, and traffic speeds will be collected by an estimated 500 induction
loop detectors, buried a few inches beneath the surface of the roads. The
project — which is known as CITRAC, for Central Integrated Traffic Control
— is due to go into operation in 1973, although much of the equipment
will be installed next year. The GLC plans to build a control centre which
will house not only the computers, but also the police supervisors who will
be responsible for taking action in the event of abnormal conditions. The
equipment will initially be installed at County Hall. (CW 13/8/70 p1)
Resurrection 90 Summer 2020 35
Forthcoming Events
Readers will, we feel sure, understand the difficulties in planning meetings
for the coming season which was scheduled to commence in September.
As things stand at the time of writing, there are no meetings planned in
Manchester and, although there is a provisional programme of London
meetings, there is considerable doubt that any face-to-face meetings can
take place in the current climate.
However, our energetic Meetings Secretary, Roger Johnson is working
towards being able to arrange events which can be attended online. As
soon as the mist clears and concrete arrangements are made, we will
publish details on our website at
www.computerconservationsociety.org/lecture.htm.
Meanwhile may we wish you all health, wealth and happiness, but most of
all health?
36 Resurrection 90 Summer 2020
Museums
SIM : Demonstrations of the replica Small-Scale Experimental Machine at
the Science and Industry Museum in Manchester are run every Tuesday,
Wednesday and Sunday between 12:00 and 14:00. Admission is free. See
www.scienceandindustrymuseum.org.uk/ for more details.
Bletchley Park : daily. Exhibition of wartime code-breaking equipment
and procedures plus tours of the wartime buildings. Go to
www.bletchleypark.org.uk to check details of times, admission charges
and special events.
The National Museum of Computing : Open Tue-Sun 10:30-17.00.
Situated on the Bletchley Park campus, TNMoC covers the development of
computing from the “rebuilt” Turing Bombe and Colossus codebreaking
machines via the Harwell Dekatron (the world’s oldest working computer)
to the present day. From ICL mainframes to hand-held computers.
Please note that TNMoC is independent of Bletchley Park Trust and there is
a separate admission charge. Visitors do not need to visit Bletchley Park
Trust to visit TNMoC. See www.tnmoc.org for more details.
Science Museum :
There is an excellent display of computing and mathematics machines on
the second floor. The Information Age gallery explores “Six Networks
which Changed the World” and includes a CDC 6600 computer and its
Russian equivalent the BESM-6 as well as Pilot ACE, arguably the world’s
third oldest surviving computer.
The new Mathematics Gallery has the Elliott 401 and the Julius Totalisater,
both of which were the subject of CCS projects in years past, and much
else besides.
Other galleries include displays ranging from ICT card-sorters to Cray
supercomputers. Admission is free. See www.sciencemuseum.org.uk for
more details.
Other Museums :
At www.computerconservationsociety.org/museums.htm can be found
brief descriptions of various UK computing museums which may be of
interest to members.
North West Group contact details Chair Bob Geatrell: Tel: 01457-868700.
Email: [email protected]
Secretary Alan Pickwick: Tel: 0161 973 6796.
Email: [email protected]
Committee of the Society
Chair: David Morriss FBCS CEng CITP: [email protected]
Secretary: Rachel Burnett FBCS CITP Hon D. Tech: [email protected]
Treasurer: Arthur Dransfield CEng FBCS CITP: [email protected]
Chair, North West Group: Bob Geatrell: [email protected]
Secretary, North West Group: Alan Pickwick MBCS FRAS: [email protected]
Resurrection Editor: Dik Leatherdale MBCS: [email protected]
Website Editor: Dik Leatherdale MBCS: [email protected]
London Meetings Secretary: Dr Roger Johnson FBCS: [email protected]
Membership Secretary: Bill Barksfield CEng MBCS CITP [email protected]
Media Officer: Dan Hayton MBCS FRSA: [email protected]
Digital Archivist: Prof. Simon Lavington FBCS FIEE CEng: [email protected]
Awards Sub-Committee Co-ordinator: Peta Walmisley: [email protected]
Awards Sub-Committee Rachel Burnett (chair), Roger Johnson
Museum Representatives Bletchley Park Trust: Peronel Craddock: [email protected]
Science Museum: Rachel Boon: [email protected]
National Museum of Computing: Kevin Murrell FBCS: [email protected]
Project Leaders SSEM: Chris Burton CEng FIEE FBCS: [email protected]
Bombe: John Harper Hon FBCS CEng MIEE: [email protected]
Elliott 8/900 Series: Terry Froggatt CEng MBCS: [email protected]
Software Conservation: Dr David Holdsworth Hon FBCS: [email protected]
ICT 1301: Rod Brown: [email protected]
Harwell Dekatron Computer: Delwyn Holroyd: [email protected]
HEC-1: Kevin Murrell FBCS: [email protected]
DEC: Kevin Murrell FBCS: [email protected]
Our Computer Heritage: Prof. Simon Lavington FBCS FIEE CEng: [email protected]
ICL 2966/1900: Delwyn Holroyd: [email protected]
Analytical Engine: Dr Doron Swade MBE FBCS: [email protected]
EDSAC: Dr Andrew Herbert OBE FREng: [email protected]
Bloodhound Missile/Argus: Peter Harry: [email protected]
IBM Group: Peter Short MBCS: [email protected]
Data Recovery: Delwyn Holroyd: [email protected]
Co-opted Members
Prof. Martin Campbell-Kelly FBCS CITP FLSW: [email protected]
Resurrection is the journal of the Computer Conservation Society.
Editor — Dik Leatherdale
Printed by — BCS the Chartered Institute for IT
© Computer Conservation Society