At Fri, 08 Apr 2005 16:52:50 -0700, Erik Nordmark wrote:
1. Add text to say "SHOULD limit packets with an anycast source to 1280 bytes". Add note that ICMP-based tools don't work with an anycast source address.
Unless, of course, the application protocol in question happens to be one of those for which, after all these years of trying to relegate IP fragmentation to the dustbin of history, we still have no better transport protocol than UDP without path MTU discovery, and if that means that we have to send fragments, so be it, we send fragments.
That's ok, as long as no packet is larger than 1280 bytes i.e. that the sender fragments to 1280 bytes instead of, say, 1500.
Yes, there are such protocols, and yes, at least one of them is being used today with anycast source addresses.
It's not that fragmented UDP without path MTU discovery is good. It's just that sometimes all the other alternatives are even worse.
Sounds like my suggestion above needs to be clarified in that it doesn't prohibit fragmentation. Sending at 1280 bytes is the IPv6 way to run without path MTU discovery i.e. the logical counterpart of not setting DF in IPv4.
Erik
-------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
