> From: "Perry E. Metzger" <perry(_at_)piermont(_dot_)com>
> Identifier/locator separation has been a topic of conversation
at
the
> IETF for at least the last decade if not longer. In spite of
this
> continuous interest, an actual fruitful proposal has yet to
arrive.
As you seem to have forgotten since the last time I pointed this out
to
you,
MobileIPv6 represents a fully-worked-out design which separates
identity
from location. (The Home Address is still used as a rendezvous point,
it
is
true.)
Mobile IPv6 can indeed be used to solve part of the host multi-homing
problem. A multi-homed host will have as many addresses as interfaces.
At the start of any given TCP connection or UDP association, one of
these addresses can be picked as the home address for the connection;
when the need appears, binding updates can be used to move the
connection or the association to different interface. We must however
solve the double jump problem: if the two hosts "move" at the same time,
the connection will only survive if at least one of the two "home
addresses" remains reachable; this will not be true if the two "home
interfaces" were to the same provider, and that provider fails. There
are a few obvious solutions; we should pick one and standardize it.
The part of host multi-homing that mobile IPv6 does not solve well is
the selection of the destination address: in the absence of external
knowledge, all destination addresses look equal, and one will be picked
at random. The destination can use a binding update to redirect the
connection to the preferred interface, but it can only do so if the
non-preferred address is reachable; if it is not, we get a
trial-and-error approach, which kind of work but is slow. There are two
known solutions to this problem: a dynamic DNS update, and the
establishment of a back-up tunnel. Both have issues, which would require
more development than a single paragraph.
It is also possible to treat site multi-homing as a variation of host
multi-homing, assigning as many addresses to each host as there are
providers to the site. I often hear the argument that this is
unrealistic, that host software will not be able to handle multiple
addresses, but if I run "ipconfig" on my desktop, I get a list of 14
addresses:
3ffe:8311:ffff:f282:8cc5:8c07:78bb:77ac
3ffe:8311:ffff:f282:69a3:a0b3:2a:d858
3ffe:8311:ffff:f282:a8e5:48c4:5a79:ec40
3ffe:8311:ffff:f282:a101:2c3f:c6bd:29b
3ffe:8311:ffff:f282:90e:f1a0:d2c7:36c0
3ffe:8311:ffff:f282:3198:8248:b889:cd74
3ffe:8311:ffff:f282:a12f:7a3a:5dbd:9eed
fec0::f282:206:5bff:fe5e:7527%1
3ffe:8311:ffff:f282:206:5bff:fe5e:7527
fe80::206:5bff:fe5e:7527%4
2002:9d3b:8849::9d3b:8849
3ffe:2900:d005:f28b:0:5efe:157.59.136.73
fec0::f28b:0:5efe:157.59.136.73%1
fe80::5efe:157.59.136.73%2
Somehow, the system manages to handle these 14 addresses; it may not
handle it perfectly, but we can certainly expect improvements over time.
So, we have at least one conceptual solution: use a variation of Mobile
IPv6 to improve host multi-homing; solve site-multi-homing by treating
it as a variation of host-multi-homing.
-- Christian Huitema