Olivier,
1. Employ something like NERD for the mapping table, no caching
By doing this, you assume that the only benefits of a mapping system
is to reduce the number of prefixes carried on core routers. In this
case, a static mapping system could be acceptable, but a mapping
system can offer many more benefits than simply improving the
scalability of the core routers. A dynamic mapping systems can allow
sites to do much more, see :
http://inl.info.ucl.ac.be/publications/evaluating-benefits-locatoridentifier
I agree that we want more than merely the reduction of the # of
prefixes. However, the question is what parts of that "more" rest on the
mapping system and what is in the responsibility of the two networks
talking to each other in an end to end fashion.
I a sense, I feel that we are talking about what information should be
cached. What I was saying above was about the ID->LOC mapping table. I
agree that a network A and network B need more information than this in
order to do, for instance, failover and traffic engineering.
But does that information have to be in the mapping system? It is
something that the two ends can also establish by themselves. In the
process of discovering this information they may some cached state. For
instance, one of the two edge routers is actually down.
This will also have similar effects as full-blown caching of all mapping
information. You might lose a packet if one of the chosen edge router
was down. You might have delay if you sent the packet to a non-preferred
router. We need to find out if the effects of this type of caching are
harmful, too. But I would expect the effects to be much less significant
than in wholesale caching. (But if the effects turn out to be harmful,
then we face a difficult problem of creating a large distributed
database that also needs to change at a fast pace.)
and we do not attempt to solve the host identifier-locator split
I think that in some environments locators would be attached only to
routers while in other environments locators would be attached to (at
least some) hosts. The architecture should support this, at least in
the long term.
I'm not sure I understand what you are saying above. Can you elaborate
this a bit?
Jari
--
to unsubscribe send a message to [EMAIL PROTECTED] with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg
