nagle@well.sf.ca.us (John Nagle) (10/07/90)
Just read "IBM's Early Computers" (Bashe et al., 1987, MIT Press, ISBN
0-262-02225-7). Some very advanced technology was developed by IBM for NSA
in the '50s and '60s, and is now coming to light. It's very impressive
what they were able to do.
One system delivered in 1962 included the Tractor tape system.
This was the first automatic tape library. The tapes themselves
were 1.75" wide, recorded at 3000 bpi and 22 tracks across, running
at 235ips. Hamming code was used for correction in each frame
(transverse, not longitudinally), so the 22 tracks actually stored two
bytes plus error correction bits. Deskew was electronic, with shift
registers, the first use of that technology. The tapes were in huge
cartridges weighing 15 lbs each, and a mechanical changer provided
access to 160 cartridges for each pair of tape drives. The entire storage
system, with 6 tape drives and a tape library, stored about 60 gigabytes.
This in 1962.
The computer to which this was connected was a version of the
7030 "Stretch" machine. Unlike the Stretch machines delivered to
scientific labs, though, this machine had no floating point unit.
Instead, it had a "streaming" command unit, called the Harvest
streaming unit. This unit offered matching, substitution by
table lookup, tallying the occurence of specified logical relationships,
and "other operations of interest to cryptanalysts". It was twice the
size of the standard Stretch machine (a monster of its day) and was
accepted by NSA in February 1962. It was used, and worked well, for the
next 14 years.
Unlike Harvest, which was a success, there was an incredibly
ambitious but unsuccessful attempt to build a machine with a
100 MIPS rate and a 100 megaword memory with 10ns access started
in 1956! This was Project Lightning. Lightning was to be based on
cyrogenic superconducting components fabricated using thin film
technology. Unfortunately, the fabrication techniques used
were not good enough to produce reliable components, and the
project ended in 1961. The final report suggested that
"cyrotron densities of several thousand per square inch can
be operated at a kilomegacycle rate." But IBM finally gave up
on cyrotrons in 1965. (This may be just as well; cyrogenic
supercomputers would be impressive, but the operational headaches,
including the liquid helium bill, would limit their utility. And this
wasn't a high-density technology, so you'd have a huge machine running in
liquid helium inside a liquid nitrogen jacket.)
John Nagle