Yes on IP, but TCP's main function is reliability - much of the mechanism of TCP (sequence numbers, acknowledgements, timeouts and retransmissions, and checksums) is all there for that. There is one aspect of internetworking (the original term - I probably should have described PUP/CHAOS/XNS as 'internetworking protocols') which _is_ crucial, though - the multi-layer address space. We'd need that even if _everything_ in the world used Ethernet frame headers. If one tried to do path selection (usually called 'routing', but I don't use that term as it can be confused with packet forwarding) using only 48-bit interface identifiers, it just wouldn't scale to the size network we have now. The ability to aggregate groups of hosts, so that a distant routing table contains only a single entry for all of them, is crucial for scaling purposes. Without that, routing tables would have to have billions (literally; add up the numbers of different kinds of end-user devices - laptops, etc) of entries. (Heck, even XNS had network numbers, precisely for this reason. Although one could build a system which has aggregatable addresses, used for path selection, but hid them from the hosts, and used an 'invisible' mapping system to translate from them to the aggregatable 'true' addresses. The LISP networking system does this, as does the 800 and inter-provider portability capability in the 'phone system - although in both cases the input and output to the mapping system have identical syntax.) Originally, IP had only two layers in the addressing - network # and 'rest', then we added a third layer with 'subnets', and finally went to a potentially multi-layer system with CIDR. (I'm not sure what ISPs are actually doing with them now - I'm now out of touch with that world.) I found Jack Haverty's message to the internet-history list about the changing nature of 'the Internet', but alas the list archives are broken at the moment, so no URL. Noel