It was now 1971, and I'd joined Bedford as 'special projects' man and dogsbody.
Unfortunately I've lost or thrown away most of my old manuals: one can't keep everything just because it may become a museum piece in 20 years. So I'll have to describe the storage as best I can, with the summary cards I still have. I didn't program it much, in fact the only 'film work' I did was to rewrite a standard film test program so that it did a proper test of the drives without snapping the films! The 803 machines were already on the way out by the time I joined the company. Six months later they were both removed.
Data was written to magnetically coated 35mm photographic filmstock, without the optical coating of course. It was perforated down each edge, and driven by sprockets. There were two tracks down the length of the film: one track being written and read in one direction. Then you reversed the film and wrote or read the other track.
The tracks had to be formatted by the makers into blocks. Each block carried a pre-recorded block number. There were 4350 characters to a block, and the transfer speed was about 5 blocks per second. One reel could hold 7 megabytes of data. I think they were made in France. The reels of film were about a foot in diameter and quite heavy. As they aged, they also became brittle. If a film snapped you were in trouble, because it could not be joined except with a butt-weld, and this had to be done by a specialist firm. A lap joint would not go through the heads. Moreover the broken ends must first be trimmed true, and so you lost some of the block. But all blocks must be the same size, so the whole film must be reformatted. You had lost all your data. Naturally you kept at least two backups of all data films.
I've seen three film 'handlers' in use, doing a long sort. This was quite fascinating, a sort of musical ballet. Not only were the drives oscillating continuously, but the processor was doing its chicken-run imitation. A small speaker on the control panel was "connected to the least significant bit* of the accumulator". As work proceeded it clucked and twittered.
You grew to love this mother-hen. With practice you could usually tell where it was in a long program: and if you had written it yourself, you could often tell what stage it had reached!
Data was being read from one film and subdivided into linked lists which were written to the second handler. This handler would hold the first phase of the sort; and writing some lists forwards on one track, others backwards on the other track. The programmer had worked out an optimisation for least winding and rewinding of film. The drive motors starting and stopping, and the tensioners all played their tunes.
Further sorting took place, and the final merged data would be written in bursts of movement to the third handler: all accompanied by the mother-hen. Now and then there was a loud hoot from the farmyard, and an operator would come and change a film, rewind a film, load another program tape etc.
The film handlers had multiple sprung tensioning arms, and there was a clutch for the supply and take-up spools. The maintenance engineer, who called monthly, was supposed to check these clutches and the arm tensions with special gauges. Even so, adjustment required skill which he didn't have a lot of. If those clutches were not correctly adjusted, the film would snap instead of reversing cleanly. An old film was kept for testing! And, as I mentioned earlier, there was a special utility program for checking the film handlers. After this snapped three films, I was asked to make it less violent in its tests - which I did. It was an interesting bit of learning for me.
A card reader, digital plotter and online line printer could be attached: but our line printer was off-line and got its data from the paper tape punched by the computer.
While rooting about in my loft, I've just found four 7" tape reels on which is a complete 'training course' for the Elliott 803: at 1+7/8"/sec ! (also several dozen symphonies). If I can persuade my ancient reel-to-reel tape recorder to work still, I may play some of this one day, and see if they've survived the heat and cold in a loft. Glad they're not on film!
(*Comment from David: Peter Onion has emailed me to mention that Gnome is mistaken here - according to logic diagrams in his possession the speaker is in fact connected to the most significant bit of the instruction register.)