Noel Chiappa wrote:
Which is why I am urging the IETF to be _realistic_ now, and accept the world
as it actually is, and set direction from here on out based on that, and not
on what we wish would happen.
The only realistic approach is to accept IPv4 at least for next
10 or 20 years, which is possible with port restricted IP while
keeping the end to end transparency.
Which means, for instance, that any design for
architecural change (e.g. introducing separation of location and identity) is
going to be somewhat ugly, because we don't have a clean sheet of paper to
work with.
ID locator separation is not essential. All we need is an architecture
to handle multiple addresses (which may be raw addresses or an ID and
multiple locators).
It also means accepting that we have multiple naming domains at
the end-end level,
It means to handle multiple IP addresses.
and will for the forseeable future;
It means to handle multiple IPv4 addresses.
and trying to work out
an architectual direction for coping with that ('get rid of it' doesn't
count). Etc, etc, etc.
The basic architectural problem so many people want to ignore
is that IP is connectionless, which means there is no time out
at the IP layer to know some address is not working.
As a result, there are a lot of wrong proposals seen in multi6
WG, which declare TCP connections are dead merely because
no traffic is observed for a while, which is no different from
poor legacy NAT violating the end to end principle.
Interestingly enough, IPv6 failed partly because its neighbor
discovery has introduced a lot of time out at the IP layer,
which is architecturally wrong (in this case, timing depends
on link layers). Requirements on RtrAdvInterval was finally
loosened in RFC3775 (MIPv6) but rest remains.
Once it is recognized that the problem of multiple addresses can
be solve only at (in case of TCP) or above (in case of UDP) the
transport layer, the solution is easy and straight forward.
Details are documented in draft-ohta-e2e-multihoming-00.txt
in Apr. 2000.
Though my experimental implementation is in IPv6 with ID/locator
separation, same is doable with (port restricted) IPv4.
Masataka Ohta
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf