I still have an uneasy feeling about the changing flow IDs across the
same TCP session. It feels wrong.
That being said, Ran's argument that different classifications for
fragmented/non-fragmented packets already happening for load-balancing
reasons (and presumably even for IPv4) when a fragmented packet hits a
backbone router and the port numbers are not in the packet.
I'm fine with Brian's suggested text. I would personally open the text
up a bit more, however. I would explain, for instance, why RFC 4311 is
relevant. I would explain the tradeoffs between packet reassembly,
different flow IDs, and using just 2-tuples.
Jari
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------