jas@MONK.PROTEON.COM (John A. Shriver) (07/14/87)
I've never seen any specification that requires IP routers to examine the Protocol field in an IP datagram being forwarded. I would argue that it is improper for a IP router to do so. This prevents consenting users of the internet to use an experimental Protocol across the Internet since some "router czar" has forbidden this protocol. My idea of IP, and IP routers, is that it should be completely blind to what protocol is in use above it (with the exception of ICMP). This is the spirit of layering. Another reason not to do this is that it's just *another* field to have to check in the main forwarding loop of a router. If everyone solves all of the Internet's "control" problems in routers "with just one little check" here or there, we'll never get the sort of performance out of routers that the Internet community seems to want. We have to keep forwarding packets as *simple* as possible, or we'll have 68030 routers running at 50 (exaggeration) packets/second. (Obviously, these are my opinions, not Proteon's...)
Mills@UDEL.EDU (07/14/87)
John, My remarks were confined strictly to the local-use issue and only when the firewall is necessary. It turns out that the fuzzballs use IP protocol 63 (decimal) for routing purposes, so they have to check that field anyway. Should it be advisable, I have no problem with this overhead in the general case. It is surely no more intrusive than the address checking suggested for generic IP gateways on this list and in recent RFCs. Dave
PERRY@VAX.DARPA.MIL (Dennis G. Perry) (07/16/87)
John, I tend to agree with you. IP routers need to be kept simple under the current architectural concepts. dennis -------