smb@ulysses.UUCP (08/09/83)
Strictly speaking, I'd say Berkeley is speaking TCP/IP; however, they're using their own link-level protocol over the Ethernet. That is, nothing above the level of the Ethernet device drivers ever sees the funny packet format, so the IP specs aren't violated. (Note also that no other part of the code sees the Ethernet header, nor the DDCMP header on the DMR-11 driver, etc.) The problem, then, is that there's no standard (that I'm aware of) for encapsulating IP packets on an Ethernet, except maybe via a type number assigned by Xerox. (What has been assigned, incidentally? I never received any answer from Xerox when I asked them.) Now -- this doesn't mean that I think Berkeley is right to use this protocol unconditionally. Yes, it's supposed to be more efficient, but I run non-Berkeley TCP/IP machines on the same Ethernet. There are other issues as well, such as the mapping from Internet addresses to Ethernet addresses. Who has blessed the Address Resolution Protocol, for example? I'll really have to hack some code to put that in when I get a Berkeley version that wants it. I suppose my point is that just saying "Ethernet" means very little past the shape of the plug and the packets; even saying "Ethernet + TCP/IP" means little more. *sigh* --Steve Bellovin