22 Sep
2004
22 Sep
'04
1:17 p.m.
On Wed, 2004-09-22 at 16:13, William Maddox wrote:
Do you know why a bi-quinary encoding was used
instead
of the more obvious 8-4-2-1 BCD encoding scheme? I've
been a bit curious about that. More generally, I've
been curious why decimal architectures held on for so
long, given the higher cost of the CPU logic. In the
older semi-electronic punch-card calculators, it
clearly simplified the logic overall, as they were
running simple hard-wired (plug-board) programs
interfacing with decimal-oriented electromechanical
input and output. I can also see where it made sense
on smaller digit-serial designs. In a large-scale
parallel CPU, however, it's not clear to me why
a binary architecture (doing binary to decimal
conversion in software) wasn't preferred right from
the start.
Biquinary has some advantages when constructed from vacuum tubes,
because tubes are so awful as logic elements (every element in a tube
has a fairly large capacitor connected between each element, in effect,
and electrons have to travel a *fraction of an inch* within the device).
(You can even get a single tube to count to N, however unreliably, using
a phantastron circuit.)
Don't take the existence of B-to-D software conversions for granted --
on slow memory (drums for instance) that alone could double or more
runtimes already measured in hours, on machines with MTBF of hundreds of
hours (early on). Besides, what else is there to do with an electronic
automatic computer but to calculate numbers? And when you are doing so
in machine code, with no what we'd call developmental software about,
it's a substantial advantage to do so in human systems. There were many
hardware solutions to B-to-D and D-to-B, some quite involved.
For the first half of the 1950's, it wasn't universally agreed that
compilers and interpreters were even a *good idea*, since so much
performance and control would be lost, etc. FORTRAN removed most of
those objections.
At some point it was obvious that decimal wasn't the right approach from
a density and facilities view, and it Went Away.
(There exist biquinary NIXIE tubes; two anodes for the 10 digits. I have
a few I think.)
There are plenty of electronic numbering systems; things you'd think are
obvious, like Binary Coded Decimal bits weighted 8, 4, 2, 1, are not
obvious, there were/are a lot of other schemes.
I'm sure actual old timers could fill in a lot here.