[ih] The netmask

[Grant Taylor]

On 1/7/25 8:06 AM, Andrew Walding via Internet-history wrote:
> So in 1981 John Postel's famous RFC 790 introduced the classful system 
> you refer to.  From today's perspective it is a simple robbing of 
> the first 4 bits of the 32 bit address to indicate what octets demarc 
> the network address boundary:

Is it really robbing the first 4 bits for all classes?

I thought it was robbing the:

first 1 bit  for class A,
first 2 bits for class B,
first 3 bits for class C, and
first 4 bits for class D (and E).

If it was really robbing 4 bits for all classes, then we could have 16 
classes.  ;-)

> This system served the internet for 12 years or so.  Not bad, not as 
> you say, not efficient.

Didn't this system just use a "netmask"?

class A had  8-bit netmask
class B had 16-bit netmask
class C had 24-bit netmask

> Then as you mention we went to a more efficient use of the bit boundary 
> with subnet and supernet masks, rather that the byte boundary that 
> the first four bits specified in RFC 790.

Wasn't the sub-netmask introduced with CIDR?  As in the part that 
subdivided the network portion that was already divided based on class?

I've run into extremely obtuse IP configurations on things designed very 
early on in CIDR's lifetime wherein the sub-net is the number of bits 
/after/ the *netmask*.  E.g. 192.0.2.0/24 (Tnet-Net-1) has a netmask of 
255.255.0.0 and a sub-netmask of 0.0.255.0.

Yes, I am thinking about OS/390's TCP/IP configuration.

In retrospect, I've seen hints of this in other systems too.  I think 
that some Cisco IOS output hints at this if you know how to squint at it 
to see the pattern.

> One minor thing I would like to correct is that the mask is only 
> communicated in routing messages, no data packets ever carry the mask.

Arguably, routing (their protocols) and security devices / software are 
the only thing that /need/ to know anything about the (sub)netmask.



-- 
Grant. . . .
unix || die