I'm going to reply to myself to expand on this with an interesting topic... The fact that Ethernet was originally designed to behave similarly to a wireless sys

I'm going to reply to myself to expand on this with an interesting topic... The fact that Ethernet was originally designed to behave similarly to a wireless system (specifically ALOHA) has important implications on the actual protocol. Essentially, layer 2* addressing in Ethernet only exists because of this design. In today's Ethernet networks, layer 2 addressing is unnecessary and it's quite possible to use only the layer 3 addresses - we could call "layer 3 switching," which is now a standard feature on commercial managed switches, an example of doing so. However, the L2/L3 addressing split has turned out to be very useful in some situations (mostly wireless networks, on which Ethernet is based!). However, it also has a lot of disadvantages, and introduces almost comically unnecessary complications to some software-defined networking (e.g. for many VXLAN setups we need to assign pretend MAC addresses to all our peers so that we can use the kernel's bridge plumbing). This is a very clear case of the now-gone history of the protocol having a huge impact on shaping how we use it today. At the same time, there exist devices that still communicate directly over Ethernet frames, something that wouldn't even be meaningfully possible had Ethernet not been designed to run on one shared media. Fortunately these devices aren't very common, because they're a real headache in a world of equipment designed to handle IP. *I always hesitate to apply OSI model labels to the IP suite, because IP was not designed according to the OSI model and does not comply with it. However, the OSI model labels are so common that it's hard not to. That said, the OSI model can quickly become more of a hindrance than a help when we start doing things like IP-over-IP.