> From: Valdis(_dot_)Kletnieks(_at_)vt(_dot_)edu
> On Thu, 26 Jul 2001 18:52:50 CDT, Jose Manuel Arronte Garcia
> <marronte(_at_)ver(_dot_)megared(_dot_)net(_dot_)mx> said:
>> If the used 48-bit addresses in lower layer protocols, why they did
>> not use 48-bit routing-enabled addressing for the Internetwork layer?
>> Not just use the very same addresses as I may have implied (I DO NOT
>> mean that).
In other words, a separate 48-bit address space? If you have two different
address spaces, used in two different places, and needing a mapping between
them, what's the particular value in having them the same size?
(Needless to say, using the original IEEE 48 address space for internetwork
adresses wouldn't work, as the addresses aren't aggregatable.)
> Third, they were doing a new design, and the old one (NCP) had a 256
> host limit.
That was the "short leader" version of the Host-Imp protocol (6 bits of IMP,
and 2 of port, IIRC - amazing what useless junk my brain contains :-), which
was obsolescent by the time TCP/IP started. The "long leader" format (which
was the only one supported by 1981) had 24 bits - (16 of IMP, and 8 of port).
> First off, 32 was probably chosen because it was a number of octets(*)
> that fit nicely into a register. ...
> I wasn't there, but I bet '4 billion will be PLENTY' was a common
> sentiment - and understanding the amount of address space wasted by
> subnetting and the eventual need for CIDR so routing tables could be
> aggregated were still a decade down the road.
Actually, some people wanted variable length adresses, and an earlier
version of TCP/IP had them. Actually, there's this awful rumor (which I
heard at the time the decision was made, so it's not an urban legend from
much later :-) that part of the reason they were removed was the limited
number of (pointer) registers available at interrupt time in the TENEX OS!
(A rumor Vint thoroughly deprecates! :-)
However, it is true that ease of processing (and even moreso, ease of coding
- the number of personnel at that point was rather limited) were big factors
in the choice of 32 bits. The growth issues were completely invisible at the
time, hence the original design allowing for only 256 networks!
(Oddly enough, although the resulting bottom 24 bits of 'rest' allowed
direct mapping of the IMP 24-bit addresses, the mapping chosen only allowed
8 bits of IMP, and had a reserved '0' byte.)
Noel