On 22 Jan 2024, at 15:40, Warren Kumari <war...@kumari.net> wrote: > So, if this already works, why are we writing a document?!
I wandered into that document expecting to find an heretical inference that control messages were not important for some reason which would have made me a bit animated, but as it turns out I quite like all of this and it does seem worth writing down. Really, I should adopt a more optimistic outlook on life. There are empty sections ready to describe the gaps in routing protocols. That's good. The document would have a much more substantial reason to exist if it included that kind of thinking. [It's not clear to me what the gaps are in IS-IS and BGP. They are both already commonly used to carry multi-family routing information; IS-IS already uses a third address family as a scaffold to hang v4 and v6 topologies on and is hence in some ways independent of either. A v4 NLRI in BGP with an IPv4 NEXT_HOP is surely valid and useable if the exit interface and layer-2 destination corresponding to that v4 next hop can be identified from a multi-family IGP and v6 neighbour discovery. Are there actually gaps in the protocols that would need to be filled before they could be used in this kind scenario? Or are the gaps related to how they are implemented, or in deployment decisions? If there was an existence proof of how to make this work with actual deployed code today, that would be quite cool.] For traceroute, could we imagine burying a v6 address in an ICMP unreachable/would fragment message somewhere? Maybe there's a reason to think that IP options could be tolerated in particular scenarios (e.g. somewhat throw-away scenarios like traceroute), or maybe that the payload in such ICMP messages is sufficently variable in practical implementation that some minor liberties might be taken? Presumably the reason to focus on carrying v4 over a network with no configured v4 interface addresses is the runout and an expectation of a slow migration to single-stack v6 for all the things. But perhaps it's also worth mentioning the reverse scenario (carrying v6 over a network with no global-scope v6 interface addresses), for completeness and to avoid a special case. Fragmentation in v6 requires in that case is a similar problem to v4 PMTUd and perhaps could enjoy similar solutions, even though there are some semantic differences between them. There are different minimum MTUs specified for v4 and v6. It seems fairly obvious that if you want to carry v4 traffic over your mainly v6 network you are fine since the v6 minimum MTU is the higher, but perhaps it's worth mentioning the thought even though it doesn't require much thinking. Joe _______________________________________________ Int-area mailing list Int-area@ietf.org https://www.ietf.org/mailman/listinfo/int-area
