This is more of a question asked out of curiosity than for practical reasons, but it's bugging me, because it might point to a lack of understanding elsewhere. Back when classful IP addresses were used, how were IP addresses translated to MAC addresses in class B and class A networks?

I can't imagine ARP being a good idea, since that broadcasts to every host on the network. Was that just not a problem because no-one actually used more than a fraction of their address space?

Thanks for taking the time to look at this.

  • ARP broadcasts are completely unrelated to the size of your IP subnet - they are limited by the size of your L2 broadcast domain. Oct 20 '18 at 11:45

From what I can gather reading old RFC/IENs

The early internet was generally built on top of existing large(ish) scale networks rather than directly on top of technologies like Ethernet. Static mappings were defined between internet addresses and the internal addresses of these existing networks . As can been seen in IEN 91 some of these networks had some concept of heirachical addressing, one even used the term "subnet".

IP over Ethernet was standardised in 1984 by RFC 894 and it seems that soon afterwards people were indeed realising about the problem you described.

In 1985 RFC 950 came out, introducing a standardized procedure for subnetting IP networks. This introduced what we now know as the "subnet mask" (though it didn't use that term), a user-specified bit mask used to seperate the "subnet address" from the "host address".

Later the concepts of classless inter-domain routing and varible length subnet masking came along. Taken together these effectively removed the distinction between a "network" and a "subnet".

  • Wow, thank you for digging into this. This certainly clears things up. Oct 19 '18 at 19:43

RFC 826, An Ethernet Address Resolution Protocol -- or -- Converting Network Protocol Addresses to 48.bit Ethernet Address for Transmission on Ethernet Hardware was created to be able to resolve IPv4 addresses to MAC addresses on a LAN. It was published in 1982, and network classes were deprecated in 1993.

Just because a network can theoretically have as many hosts as Class A or Class B networks allow doesn't mean that it actually has that many hosts. Consider that in 1982, ethernet was on coax, limited to 500 meters, it really wasn't possible to get anywhere near that many hosts on a single LAN.

  • So, just to make sure I understand your answer: ARP was actually used in class A networks, too, but the networks were just very sparsely populated? Oct 19 '18 at 16:04
  • Yes. A Class A network would be assigned to a company, but the company would subnet the Class A to smaller networks, and ARP would be used on those networks. Remember that ARP is a broadcast on a LAN, not across routers. The routers inside the company would bound ARP.
    – Ron Maupin
    Oct 19 '18 at 16:09
  • 1
    Back in the days of classes, there was no "subnet the Class A". Yes, companies put hundreds / thousands of machines in the same broadcast domain. And yes, it worked about as well as you might imagine. (these were also the Dark Times™ of routeless proxy-arp)
    – Ricky
    Oct 19 '18 at 16:20
  • Actually, you could subnet a Class A network, although the subnets all had to be the same size. It was in Inter-Domain routing where you could not subnet a classful network. Within a company, you could subnet a classful network.
    – Ron Maupin
    Oct 19 '18 at 16:25

In those pre-CIDR days, a lot of equipment already had configurable netmasks, which were often described as "subnetting", and many networks just considered the classful masks as defaults which weren't used.

Ethernet was still fighting for supremacy, and the issue of collisions and per-packet interrupts were considered by many to be downsides of the technology. Others said they weren't important or would be obviated by later hardware and that the network effect on costs of ubiquitous compatible network hardware would be overwhelming.

A great contemporary paper is "Measured capacity of an Ethernet: myths and reality", D. R. Boggs, J. C. Mogul, C. A. Kent. ACM SIGCOMM Computer Communication Review, Volume 18 Issue 4, August 1988, Pages 222-234. You'll easily find it online. (Also reprinted in the "25th anniversity" edition of 1995.)

It describes the size of ethernet networks then as "tens to hundreds".

  • That all fails apart with the one machine that can't have it's netmask altered. (and there always was one!) Or routing protocols that didn't share prefix lengths, so the entire network has to be the same. (jumbo frame is the same class of problem)
    – Ricky
    Oct 19 '18 at 16:24
  • Of course! It was a bodge, and CIDR was obvious in hindsight. Also horrible were vampire taps and long, thin, single points of failure which went through every inaccessible duct in the building.
    – jonathanjo
    Oct 19 '18 at 16:27
  • Many thanks to both of you. Interesting stuff. Oct 19 '18 at 19:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.