On Aug 10, 2015, at 10:28, Fred Baker (fred) <f...@cisco.com> wrote:
>
> If every router is responsible to announce prefixes from ISP-Alice and
> ISP-Bob on every LAN, then Spanky has a distinct probability that, to get a
> packet to ISP-Alice, it will send it to ISP-Bob, who will then have to
> redirect it to ISP-Alice. If on that LAN, Alice advertises the ISp-Alice
> prefix and Bob advertises the ISP-Bob prefix, and Spanky presents the packet
> to the router that advertised its source prefix, Spanky will invariably
> present such a packet to Alice.
To send a packet through ISP-Alice, it suffices for Spanky to send to any
default router on its link that has a source-specific route for ISP-Alice
through Alice, which HNCP seems to be capable of insuring that Bob will have if
it’s smart enough to be advertising in its RA messages a prefix assigned from
one of Alice’s delegations.
Admittedly, this might not be optimal— a direct path from Spanky through Alice
to ISP-Alice might perform better, as the section on Residual Issues in the
draft notes— but there isn’t any obvious way currently defined to signal to
hosts that a particular default router should be preferred over others for
packets sourced from addresses corresponding to a Prefix Information Option. It
may be tempting to think that conformance with RFC 6724 could help in the case
where routers are coordinating to advertise only the assigned prefixes for
which they are currently the best default router, but I suspect that it isn’t
so simple and serious complications involving topology reconfiguration and RA
timeouts can arise.
I think that’s a general problem not specific to HOMENET, and 6MAN should
decide what to think about I-D.baker-6man-multi-homed-host accordingly.
—james
_______________________________________________
homenet mailing list
homenet@ietf.org
https://www.ietf.org/mailman/listinfo/homenet
