I don't think those are the only two alternatives. For example,
something like the APT idea where the full mappings are shipped to a
set of devices such that such a full set is near to and on the query
path of any leaf is an alternative worth examining.
We have stated before that LISP-ALT routers could cache mappings. This
can help with Map-Request latency. And if the mappings are cached do
to seeing Map-Replies or because they are pushed with some other
protocol, then so be it.
From a design perspective therefore, I would tend to look for a
system where a device can be sent unsolicited mappings, can choose
to keep them, and can issue a query via an overlay to resolve
mappings it does not have.
They would have to be signed.
While one could argue that this complicates the protocol, it is an
unneeded complication only if we are very sure we know what the
right answer is for information distribution. It is admittedly more
complex than just using BGP to carry EIDs. Are we sufficiently sure?
Agree.
It then becomes an operational decision whether the mappings are in
any given leaf, in some intermediate devices, or in just the
originating leaves. (While the decision on the alternatives can not
be made wholly independently by all the branches of the overlay
tree, there is room for flexibility.)
Right, but is a state/latency tradeoff.
Dino
Yours,
Joel
Dino Farinacci wrote:
Dino -
...
The point here is begging a single question:
1) Should all the mappings in the universe be in an ITR?
2) Should only the mappings for sites the ITR is currently
talking to be in the ITR?
This is the important matter. Decide on that then we can talk about
how to get the mappings where they need to be.
_______________________________________________
lisp mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lisp
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
