Re: [RAM] Incremental Deployment of LISP

[Eliot Lear <lear at cisco.com>]

Roland Dobbins wrote:
> On Mar 29, 2007, at 12:28 PM, Noel Chiappa wrote:
>
> > It's a general law of system design that as systems grow, it's 
> > usually the
> > case that a single mechanism A that did both X and Y needs to get 
> > separated
> > into two separate mechanisms B and C, one each for X and Y - which 
> > means that
> > it would be an extremely bad idea to go in the reverse direction as a 
> > system
> > got larger.
> >
> > In other words, trying to do both:
> > - finding out where some user/computer/etc is, and
> > - finding out if/how to get there
> > in a single mechanism would be a really bad idea.
>
> The DNS won't work, and deploying some other infrastructure (X.500?  
> Something Else?) to handle lookups is a nonstarter, IMHO.

Lookups are a non-starter because of the lookup latency and what you do 
with a packet on a cache miss (it's important to say why they're a 
non-starter, Roland ;-).  We want something that is akin to the RADB 
(perhaps a big table that can be diffed periodically), but to be sure is 
administered easily by end systems.  This goes to my earlier question of 
whether size matters.  Can a router keep this table in fast memory?  
OTOH, there's an argument to be made that with only PA addresses in the 
routing table you can "borrow" space from today's PI entries.

Eliot

Eliot

_______________________________________________
RAM mailing list
RAM at iab.org
https://www1.ietf.org/mailman/listinfo/ram