[Taqwus]

Another reason for octets --

Load/store instructions are often constrained to 1) operate only on data words or double words, and 2) operate only at *word boundaries* -- e.g. for 32-bit words, only at addresses divisible by 4 (bytes). I'm not familiar with x86 machine language, so I can't be sure.

Ditto for arithmetic. ALUs are designed for a given number of bits in their input -- e.g. handling 32-bit numbers, and outputting 32 bits plus various flags that can be set.

So while one *could* use, say, 14-bit foo, there'll need to be instructions converting back and forth between 14-bit and, say, 32-bit, if you want to do arithmetic. Those 14 bits may need to be shifted and zero-padded to fit into a 32-bit word for arithmetic, memory usage... and depending on language support, the programmer may need to explicitly specify the shifts and other bit operations. It's much more convenient normally to rely on "standard" data types.

(Standard in quotes. C, for instance, specifies only minimum ranges and the ordering -- shorts must be at least 16 bits, and no longer than ints, but other than that could pretty be whatever. Endianness is also not specified. Java is more standardized IIRC -- fixed bit lengths, and network-byte order.)

------------------
-- The thing that goes bump in the night