RFC1883 and ipv6 spec v2

[Matt Crawford]

> It's clear that there are two camps on the flow ID - those who
> believe it should be used for label switching, in which case
> it changes on the fly, and those who believe it should be an
> invariant. I agree we need a resolution of this but there are
> arguments both ways. Anybody want to do the pro/con analysis?

For invariant: IPv4 will be around for a long while.  Using the IPv6
flow id for label switching buys nothing for IPv4.  Putting the label
at a lower layer (e.g., layer "2.5") works equally for all protocols.


			       *  *  *

Here's an efficiency hack, for free.  Include a byte in the lower
layer label which is initalized to the TTL or Hop Limit, decremented
at each label-switched hop (by having the outgoing label have a value
one less in that byte), and copied back to the TTL (with checksum
update) or HL field of the header when the packet is forwarded to a
non-label-switched link, or when an ICMP error is to be generated for
the packet.