Being safe on the internet (was Re: Here we go again - ISP DPI, but is it interception?)
pwt at iosis.co.uk
Tue Aug 10 17:27:54 BST 2010
Tom Thomson wrote:
> Peter Tomlinson wrote:
>> The memory stirs, taking me back to 1968 when I designed the very simple
>> memory management hardware for the ICL 1904A (and in the process fixed
>> an error in the 1906A's MMU). Took the software people another 2 years
>> to get George 4 running. So that was old-fashioned, was it, Tom? It was
>> state of the art then, in the commercial environment that soon after
>> took a wrong turn...
> No, I don't think the 1904A memory management system was old-fashioned in 1968, although machines with safe memory management had by then been around for at least 5 years, for example both versions of Atlas antedated it and were in some respects more advanced (as was the memory management on its contemporary English Electric 4-75) and at least some 360s that predated it had decently secure memory management. The disasters didn't happen until quite a few years later when the segmented memory and hard distinction between programme and data of early mini-processors (such as the CTL Modular 1) went out of fashion, and new fashions decreed that memory management systems like those were the province only of that unfashionable creature "the mainframe", ICL with its 1900, System 4, and 2900 ranges, Burroughs, IBM and so on were all makers of those unfashionable mainframes, academic projects like Multics and MU5 were not trendy because they insisted on trying to do verifiable security somewhere near right, and even Intel (quite a bit more later) with its segmented memory was undesirable (until MS decided to completely bypass the memory protection) compared to the fashionable exploitation of cheap hardware using cheap and sloppy programming standards.
> Was it a commercial pressure that brought us to a wrong turn, or was it the flight of the crowd towards the fashionable? My feeling is the latter - I saw the pressures to follow fashion with good money being thrown after bad at doomed projects because they were thought to be fashionable, you must have seen that too.
In my view it was first the merger creating ICL that caused a loss of
the ICT determination to keep on forging ahead and succeeding
financially. And, despite the problems of getting the 1906A finished,
plus the delays to first George 3 and then 4, before the merger we were
hearing of worse problems at Kidsgrove. Engineers who were supposed to
be designing the 4-85 being diverted to 4-70 / 4-75... 
But I left in autumn 68, although I spent the following summer doing a
little work on NR ideas.
NR should, in my view, have had a soft microcode, i.e. stored in
rewriteable memory. Instead it seemed to plod on with hard coding (maybe
someone can confirm or contradict my assumption). And then the
engineering discipline that had dominated West Gorton failed in the
combined company - that was the real commercial wrong turn. Eventually
the Chinese Army syndrome set in. I remember visiting West Gorton when
there was the first small batch of 2970s being commissioned - by chance
Alan Chambers and I, in the area for a meeting at Daresbury, just walked
in through the door, asked for Rob, and he said to come in - the
shutters had only gone up that week . He had worked under me in my
last 6 months at West Gorton, and hinted at a certain amount of chaos as
we gazed at a 2970 doing nothing. In 67/68 we built one 1904A prototype,
made it work (very quickly), then it went to manufacture, but it seemed
that NR went into pre-production before it was ready. Later I remember
visiting the 2980 installed at Bath , seeing the Chinese Army in
action, testing (sic) software that didn't work properly.
 Floating Point Unit was specifically mentioned (and later I heard
that its error distribution was asymmetrical, causing problems with
scientific libraries - it probably needed an extra bit or 2 in the FP
unit). Curiously the FPU was the most problematic part of the 1904A, the
small team working on that having deigned to use Charles Lindsey's
Simbol simulation system running on the Manchester Atlas (as I have
noted before, I wrote the CPU simulation model).
 That we had been let in caused a certain amount of fluttering in a
dovecote elsewhere, and led to a very interesting encounter some time later.
 The South West Regional Centre - again management mistakes as well,
with almost all of the Bath team recruited from outside rather than from
the 5 client HE institutions who had been working closely together for
More information about the ukcrypto