Character width; was: these RTL or what?

[Chuck Guzis]

On 6 Oct 2007 at 21:52, William Donzelli wrote:

> Six bits also is VERY inefficient for numeric data - all that stuff
> the bank uses. In the 1960s, IBM found this was a HUGE percentage of
> what was stored on disk. Yes, you could pack 1.5 BCD digits into six
> bits, but then there is a overhead dealing with the translation. The
> overhead may be tiny on a per digit basis, but when you need to crunch
> zillions of digits per nightly run of the batch...

If efficiency was the primary goal, hollerith cards could have been 
punched in column-  or row-binary; packing 3 digits per colum, with 
leftover codes for sign and point.  Think of the trees saved!  Yet I 
never saw a BDP application do that.  Would have been hell on the 
punches in any case.

> > In retrospect, a 12-bit character would have made a huge amount of
> > sense--12 rows on a punch card, so nn need for translation hardware
> > to somehow turn legal and illegal punch combinations into characters.
> 
> Once again, very inefficient for character data - at least character
> data back then. Too many wasted bits.

Back to 6 bits.  For the type of stuff that the BDP community did, 6 
bits per character was plenty.   

In that sense, using an 8 bit character was a real waste.   Two 6-bit 
characters can pack 3 BCD digits--exactly the same efficiency as 
S/360 packed decimal--and packing/unpacking is not much different 
from converting from packed to zoned decimal.

Cheers,
Chuck