I support this document. As a general statement, ND has survived for 15 years with relatively few changes from the original specification. We can all take pride in that success.
That said, a number of the individual decisions made at that time were somewhat arbitrary, and reflected a view of the world consisting of ethernets (and other LAN technologies) defined by the state-of-the-art available at that time. As systems and technologies have evolved in the last 15 years, some of the boundaries of the ND spec seem worthy of revisiting to take into account what actual deployments are doing today (or will be in the near future). As one example, the cost of multicast on wifi is significantly higher than unicast. Yet, the ND spec calls for deleting an NCE if NUD fails, resulting in a fallback to multicast for address resolution. While this is a simple and straightforward way to make the protocol work, I can imagine worthwhile optimizations whereby ND is more conservative in falling back to multicast and deleting an NCE that was working fine just a few seconds ago. For example, one could continue probing a neighbor using both unicast and multicast, and one could continue to forward traffic to a unicast neighbor vs. discarding it because the NCE has been deleted. I'm sure there are other examples where tweaks to the basic algorithms would be worth considering (so long as they are justified be real operational concerns). Thomas -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
