tcp/ip Mystery

[spatio]

I have a text which indicates that a Class A network, using the 1st 8 bits for network id, can have 126 Networks, and 16,777,214 hosts. In Class A, the first bit is 0, which leaves all remaining combinations of the rest of the 7 bits for uniqe NetworkID's. 2 to the 7th power is 128 combinations, with decimal values 0-127. So why can you apparently only have 126 NetworksID's within a Class A Address. Also 2 to the 24th power is 16,777,216 for the possible combinations of the remaining 24 bits for unique host ID's. Again there seems to be a discrepency of 2. I can see where the decimal 0 might be eliminated, but then I am still left w/ a discrepency of 1. Can anyone shed some light on this mystery?

thanks
Spatio

 

[drebo]

There's a rule in classful networking that says you cannot use the first available network nor the last available network. Just like you can't use the first host (network address) or the last host (broadcast address.)

However, classful networking doesn't exist, so it's all irrelevant.

I've been Cisco certified (CCNA and CCNP) for a decade now and never knew until recently that Class A addresses are those whose first bit is 0, Class B addresses first two bits are 10, class C addresses first two bits are 110, etc. Why didn't I know this? Because you don't need to. Not for Cisco certifications and not for practical applications.

The only people who need to concern themselves with classful networking are historians, and even that's debatable because classful networking was never actually implemented.

 

[spatio]

that was interesting,
what more can i say?

thanks again,
spatio

 

[Gryz]

because classful networking was never actually implemented.

For historical reasons: yes, it was implemented.

RIPv1 does class-full routing only. The RIP-packet format does not have any fields to describe the prefix-length (or network mask).

Later implementations would assume that all subnets in a class-full network would have the same prefix-length (and thus the same subnet-length and host-part length). A router would deduct the subnet-length from an example that is configured on its own interfaces.

BGPv1 - BGPv3 were class-full. Again, there were no fields in the update packets to describe prefix-length or subnet mask. BGPv4 changed all that, and introduced prefix-lengths. BGPv4 was deployed somewhere around 1993-1994 (in cisco's IOS 9.21). That's when CIDR really started.


But you are absolutely right of course. Anyone doing networking these days should forget about class-full routing. It confuses people a lot more than it helps them understand things.


Networks get assigned a prefix.
Prefixes are bit-strings 1-32 bits.
Hosts get assigned an IP address.
Addresses are exactly 32 bits.
Hosts connected to a network get assigned an IP address from the prefix of the network.

That's all you need to know.

 

[iamgenius]

Yet you see it in the wiki page:

http://en.wikipedia.org/wiki/IP_address

 

[imagoon]

Wiki isn't exactly authoritative and those who know IP and IP routing will never bother to look at it. That part the drebo mentioned really is meaning less today so why bother remembering it etc.

Now "Class a,b,c" generally refer to the subnet mask length and pretty loosely. Knowing the old 0.0.0.0 -> 127.255.255.255 thing doesn't really help any more. Knowing how to subnet all the range and the exceptions for each range helps more.

 

[Qrilock]

Beyond the deprecated classfull rules for only having 126 networks in the 'Class A' range there are 2 special networks. 0.0.0.0/8 is reserved for special use as broadcast to current network and is only valid as a source address. 127.0.0.0/8 is reserved for special use as loopback/localhost addressing. Neither of these is usable as a subnet.