[ppml] Revising Centrally Assigned ULA draft

[David Williamson]

On Mon, Jun 18, 2007 at 11:32:47AM -0700, David Conrad wrote:
> Is it just me or is there appears to be a disconnect here:
> 
> - On the one hand, we're told that routers can today handle an order  
> of magnitude more "routing slots" and over the long term, can handle  
> oodles more and that the RIRs don't do "routability".

There's two pieces here that are interesting.  There's sufficient raw
copmuting power and space to hold a complete 10M-route table, which I
do believe routers are likely to have "soon".  The other piece is more
dynamic - you need to have some way to send/receive full table updates
in a timely fashion, and process them to convergence before the next
set of large disturbances to your local table.

Given the way BGP works, I think the latter is unlikely to be an
operable possibility at 10M-routes.  For my org, all failures would be
handled in ~100ms.  BGP presently takes minutes to converge.  Changing
that to 10s of minutes (while additional updates continue to show up)
is just not a workable solution.

The number of routing slots really is important, regrettably.  That
whole loc/id split thing will need to get solved, somehow.  I know I'm
really not looking forward to the BGP replacement period.  I suspect it
will be at least as painful as the IPv4 -> IPv6 transition.  I simply
don't think the current combination of BGP and IP is going to scale
in the way that some folks claim.  I could be wrong, but....

-David