The fascinating thing to me is the way in which the media is a hybrid of a punched card and a floppy.
It's more like a floppy - especially in its original version, the IBM 2321 Data Cell Drive of 1964 (see below). MagCards are a downscale from the 2321, which was, at its time the top end of random accessible online storage. They provide much larger storage than any punch card, and were introduced (1969) way before floppies in the early 1970s (*1).
It seems to have existed when both of those were in use, but it also seems to have been a niche solution only found in this IBM word processing use-case.
The MagCards were specifically developed for this niche as an upgrade to the former tape system.
So, it did not find widespread use as a substitute for either punched card or floppy, as far as I know.
It only utilizes the form factor of punch cards to allow the use of existing mechanics. Beside that there are no similarities with punch cards. It was developed as an upgrade to the Magnetic Tape Selectric Typewriter (MSTS) introduced in 1964 (*2).
Unbeknownst to most, IBM was a big player at dictating equipment during the 1960s. It was a whole system called "Executary". The Office Division developed several products from that base, turning into the MSTS in 1964. It can be seen as kind of a modernized Flexowriter, using a Selectric typewriter and special magnetic tape cassettes. The Tape was transported with a sprocket, much like paper tapes. While tapes could also be transferred to a mainframe, using an IBM 2495 Tape Cartridge Reader, it was mainly meant to support editing capabilities for putting dictations down to paper.
Think about the classic office circle of some cigarette smoking 1960s manager using an IBM Dictaphone 214 (*3), calling his secretary to take the tape, turn it into writing, having the printout returned, to be redacted. And that's where storing it onto tape became a thing to speed up the cycle. An equally important use was duplication - having one paper typed ('printed') several times for distribution. No more carbon paper, every copy looks like a hand made original - or if used, even more copies could be done.
It was upgraded in 1969 with the introduction of the MagCard as Magnetic Card Selectric Typewriter (MCST). One card could store about 5000 characters - advertised as 'more than a page'. It could handle a single card at a time.
Another update happened in 1973 with the MagCard II, like shown in the video you linked. Now not only was capacity increased to 8000 characters per card, but the mechanics could also handle a stack of up to 50 cards. Now even quite large papers could be handled - more importantly printed - within a (comparably) short time.
Just, by the mid 1970s, IBM lost its leadership in this area. Other manufacturers like Wang or Lexitron had taken over - all to be put aside when microprocessor-based systems entered the market around that time.
I'd like to know how the drive functioned,
The first MagCard recorded data in 50 tracks of 100 characters each. Encoding was done neither in ASCII nor EBCDIC, but kind of a Selectrics ball position code (*4)
and why its application seems to have been limited to this one niche.
Because it was only meant for that niche - and soon to be replaced by floppies during the 1970s.
IBM 2321 Data Cell Drive
IBM's Data Cell drive, as well as similar systems from competitors (including CDC as major OEM manufacturer) grew out of the idea of drum storage, but with exchangeable media. Instead of having a fixed magnetic surface, the drum featured tiny holes and a vacuum, able to suck exchangeable magnetic cards to its surface. Cards were stored in exchangeable boxes, called 'cells', each holding 200 stripes. 10 boxes could be held at a time, resulting in 381 MiB online capacity.
Access time, of course, varied a lot between reading from a loaded card and having to fetch one. Much like drums, there were several fixed heads (20). Access was instant. Unlike drums these heads could be moved into 5 positions, thus accessing 100 tracks (as groups of 20) (*5). Now access time was near 100 ms. Quite in line with the 2311 disk drive of the same time (7 MiB). This would drop to something like 600 ms, when the card addressed needs to be fetched from its cell. Still great.
Especially when considering that a cell drive held about 50+ times the capacity of said disk. Or the same as a full size 2401 tape (~45 MiB). So a 2321 held an equivalent of 10 mounted tapes with an access speed close to a disk drive. Handling wise each cell could be seen like a disk pack, and exchanged as easily - just with 5 times as much storage.
It gets even bigger, as up to 8 data cell drives could be operated from one controller, allowing instant access to three whole gigabytes.
The technology wasn't just used by IBM, but others as well:
- SIEMENS 568-11 (a CDC OEM drive) with 511 MiB at 500 ms (average)
- NCR 353/5 with 5 MiB at 110 ms
- Bull Bullrac with 324 MiB at 215 ms
(Access time always average)
As usual, all the competition used to offer either more storage, or faster access - and always faster data transmission (not noted). All of them also used wider cards than IBM did, as well as a kind of mechanical card addressing using a kind of variation of the Edge-Notched-Card. All of this was to avoid IBM patents, as IBM had patents on the way the stripe was pulled and transported, as well as how it gets reinserted in exactly the location it was taken out.
These drives were an extremely important storage solution for about a decade between 1965 and 1975. At that time a single 3336 Disk stack held ~200 MiB and a unit could hold two of them - the equivalent of a single 2321 unit.
*1 - While the floppy drive was created in 1967 as a microcode loader for mainframes, it wasn't introduced until 1971. Usage outside microcode storage was only reached after 1972 when Shugart/Memorex introduced the 650 drive. See History of the floppy disk on Wikipedia.
*2 - There is an incredible, almost expressionist, 1967 film about the system: Paperwork Explosion.
*3 - It's the hypermodern 1960s, so no more lap sitting dictation.
*4 - Ball position is encoded with rotate as -5..+5, tilt as 0..3 plus Corr, resulting in 88 valid positions encoded in 7 bit. Additional functions filled (some) holes to make it an encoding unlike any other. For example switching between upper/lower case (shift) is done with separate codes, much like in CCITT No.2. Others were index functions, tab or alike.
*5 - For synchronizing/positioning there was a 21st head reading the 101st track with sector information.