Hi,
I've been aware of this draft for a while, and have begrudgingly felt
that if any form of address translation is going to occur in IPv6 then
the method described in this I-D was a good way to do it, as it avoids
many of the drawbacks of many of the IPv4 NAT and NAPT methods.
One area where I think there could be further discussion is the
consequences of end-nodes not knowing their global address(es). There
is mention about applications having trouble because of not being able
to do referrals, as that is a well known problem in commonly deployed
IPv4 NAPT. One of the high barriers in IPv4 NAPT for this to occur was
the absence of a 1-to-1 mapping between external and internal
addresses, as well as the absence of knowledge of which external ports
are currently mapped to internal ports and vice-versa. This draft
removes both of these significant limitations in IPv6 translation.
I've only recently realised that what is really happening with address
translation is that end-nodes lose their ability to know their global
identity at the network layer (losing their ability to know their
global addresses is the symptom, losing their ability to know their
"global identity" is the consequence). If two end-nodes don't have
global identities, they are forced to communicate, at least initially,
via an intermediary which does have a global identity. End-nodes behind
translators can't have a true peer-to-peer relationship at the network
layer or above it if they don't know their own global identities. Yet my
understanding is that a true peer-to-peer property of the network layer
is one of the fundamental architectural design goals of the Internet's
network layer protocols. While IPv4 NAT/NAPT limited that significantly,
my hope has been that IPv6 restore that the network layer's true
peer-to-peer nature, which requires end-nodes to know their own global
identity.
I think two other cases could be mentioned were trading of
global identities are occurring at the transport layer, rather
than the application layer - SCTP and multipath-TCP - and that their
reliability and performance goals will also be effected by IPv6
prefix translation.
There also might be one possible saving grace worth mentioning. If
protocols and mechanisms to separate the locator and identifier become
wide spread, then I think true end-node peer-to-peer communications
should return to being commonly available.
Apologies for the late comments, I've only recently been thinking a bit
more about "global identities" after seeing SCTP behavior in more
detail as well as a IPv6 sip client incorrectly picking a ULA address
instead of a global address for an incoming media stream.
Best regards,
Mark.
(Please CC me, I'm not subscribed. Thanks.)
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf