Talk:Titan (1963 computer)

Computing Mid‑importance

	This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.ComputingWikipedia:WikiProject ComputingTemplate:WikiProject ComputingComputing articles
Mid	This article has been rated as Mid-importance on the project's importance scale.
	This article is supported by Early computers task force (assessed as High-importance).

University of Cambridge

This article is within the scope of the University of Cambridge WikiProject, an attempt to improve articles relating to the University of Cambridge, and to standardize and extend the coverage of the University in the encyclopedia. If you would like to participate, you can help us by editing the article attached to this notice, or you could visit the project page, where you can join the project, learn more about it, see what needs to be done, or contribute to the discussion.University of CambridgeWikipedia:WikiProject University of CambridgeTemplate:WikiProject University of CambridgeUniversity of Cambridge articles

Early comments

(The Titan main memory consisted of 128K of ferrite core store rather than part core part rotating drum-store as used on the Manchester Atlas)

The crucial distinction in memory organisation is surely the real vs virtual one. Virtual memory inevitably means main memory (be it core or semiconductor) backed by drum (then) or disc (now). I'm not clear what the purpose is of adding the remarks quoted above. 20:29, 4 May 2005 (UTC) Geoff97 20:33, 4 May 2005 (UTC)[reply]

Well at the time the problem was presented as one of LATENCY and no-one (in my undergraduate hearing) referred to it as virtual memory. Disks in 1960s had their own problems...

Also, I think you're viewing this too much from a machine architecture rather than a hardware perspective - even as late as the early 70s 4-wire core planes were hand-wired and cost the earth

To develop this point - this was early 60s, only a decade beyond the use of CRTs and mercury delay lines for storage. As I recall, Core was available but still expensive. Hardware was still of the essence!

The drum store was (I understood) usually programmed as main memory, in machine code where a skilled programmer could make use of the _cyclical_ fast-access to minimise latency (not do-able on a disk because of the unknown disk-head latency).

What was the cache on the Titan? There were certainly 4 'fast-registers' (Schottky diodes?) (which were often unreliable and so disabled). The core-RAM was in four banks and the fetches interleaved - as a result of this you couldn't modify a machine code order less than 4 in front of the instruction pointer, because it was already being fetched and into the decode queue.Linuxlad 15:48, 15 May 2005 (UTC)[reply]

I'm still not convinced about your emphasis on the paging mechanism on Titan - look at the www.cl.cam.ac.uk site for a discusion of old Titan cards (including a core plane) - where it makes it clear that the Titan was cut-down compared with the Manchester Atlas, and one of the units which was discarded or at least heavily modified was the paging controll. Linuxlad 21:55, 24 May 2005 (UTC)[reply]

As I put in the article, the Titan had a real (but cached) memory system, as opposed to the Atlas that had a virtual (paged) memory system. So there was no statement that there was paging memory on the Titan - quite the reverse in fact.

Regarding the cache on the Titan (which was called slave memory at the time), I do not recall the hardware details of it. However if you look at this reference: [1] where the Wilkes 1965 paper describing slave memory is refered to you will see that there is a remark to the effect that slave memory was implemented on the Titan.

A further reference for information on the Atlas one-level store (as it was called at the time, before the term virtual memory came into use) is the paper by P.J.Denning, Before Memory was Virtual: [2]

Hope this helps. Geoff97 18:25, 25 May 2005 (UTC)[reply]

Link to reference [2] updated Shannock9 (talk) 23:44, 25 March 2010 (UTC)[reply]

What doesn't seem to have been mentioned yet is that, although it was not virtual memory, it did use datum and limit registers, to provide isolation between application processes. Segments had to be on a power of two boundary corresponding to their size. I think the Titan Supervisor could also swap out user processes.

It also used memory mapped I/O.

Also, I seem to remember, that the extracode (system call) jump table was implemented as dioed matrix ROM, although I'm not so sure about that. Although I call it a jump table, many extracodes were actually implemented inline within it!

-- (not in history because logins don't take for long enough) David Woolley 19:03, 28 October 2005 (UTC)[reply]

How many?

The article says two were built with no citation, while this says there were three: a third one in Aldermaston? I will change to match the source. Perhaps that third one never was delivered? W Nowicki (talk) 21:19, 22 April 2011 (UTC)[reply]

Are more references really needed

It seems to me this page is now up to scratch.

Can an editor either take the notice off the page or be more specific about the changes they would like to see (for example the sort of citations). John Irving Tait (talk) 15:44, 6 April 2013 (UTC)JIT[reply]