Alex's post in Linux the other day led me to spending a good number of hours looking up IBM mainframe history sites - seeing what I missed as the mainframe years speed by. The more I surfed, the more respect I gained for IBM's hardware engineering.
The partcular item that caught my eye (and my interest) was this: The [link|http://www.columbia.edu/acis/history/datacell.html|IBM 2321 Data Cell]. An amazing contraption, circa 1967, it was a curious mixture of tape and random-accessability. This was basically achieved by splitting tape into hundreds of little slivers, putting them in boxes, then grabbing the slivers and whizzing them around read/write heads as and when required. Sure, ultimately it proved to be a folly - plagued by reliability issues (mostly caused by difficulty getting the slivers of tape back into their holders), but back in the day, it was a noble effort - giving (up to) gigs of random-access storage when hard-drives only offered a few MB.
Another link and bit of history [link|http://members.optushome.com.au/intaretro/2321DCD.htm|here].
Ah, the good old days - when innovation didn't have inverted commas...