I think that the only real world problem that I see today is the load
balancing. We should fix that.
Its hard to imagine that we'd really need full 20 bits for that. We
could even say use the last 16 bits for load balancing regardless of
what we decide to do about formal reserved/not reserved status for the bits.
I think the rest of the debate is rather academic. In practice the flow
label is today reserved, no matter what the RFCs say. Nothing will
happen with the flow labels unless the IETF makes a decision that we
should do <some new flow label scheme> or <some other cool thing>. And
I'm not particularly optimistic about inventing that cool thing that
would be a good match for using the remaining bits. I can see some
danger of adopting some kludges or support for something relatively
unimportant if we have the free bits, however. That being said, I
support the idea of documenting that the bits are reserved because that
does seem to match reality.
Jari
P.S. I'm not sure I understand why draft-carpenter-flow-ecmp requires
the use of the flow label to happen only in routers that have
participated in setting up the flow state... that seems like a pretty
big non-starter to me.
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------