> From: Russ White <[email protected]>
> the point is that while loc/id split might be neat ... I don't know
> that loc/id separation "solves" any sort of "table size" issue
I can give you several examples where it can reduce the table size.
First, let's consider an 'optimal entry router' problem: i.e. site X has a
large interal network, is connected to the Internet in several places, and
wants inbound traffic to take the interconnect which is 'closest' to the
destination, internally. Today, one has to do this by leaking internal routes
outside site X, i.e. not advertizing just a single route to X, outside of X.
If one has a location/identity system which i) allows giving a site border
router as the location, ii) is demand driver (i.e. mappings are not flooded),
and also allows giving different mappings to different mapping requestors,
one can fairly easily cause traffic to take the optimal entry router, by
setting the appropriate border router in the mapping reply. This obviates the
need to leak 'more-specific' routes outside a site.
Second, consider a site that wishes to multihome. Right now, the only ways to
do this are either i) give each host multiple addresses (aka SHIM6, which
most sites do not like), or ii) use PI addresses - which increases the
routing table size. In other words, the only practical method for site
multihoming currently increases the routing table size. With a
location/identity separation system in place, as long as it's one that allows
1->many mappings for the {identity -> location} binding, it's easy to
multihome a site: the mapping just gives the multiple locations which a site
is multi-homed to. This obviates the need to advertize a PI address
network-wide in order to multi-home a small site.
That's just off the top of my head, quickly - there may be other cases too.
Having pointed that out, location/identity separation is not a panacea for a
more advanced routing architecture. It can do _some_ things fairly well (e.g.
support small-site multi-homing), because one has two names available to play
with, each of which has simpler semantics than classic 'addresses', and any
time one has a binding layer, one can get some power out of that (c.f.
Lampson's Law, 'one more layer of indirection'). But it's clearly not routing.
On the other hand, many people who have looked at advanced routing
architectures do feel that location/identity separation is a necessary
precursor, for a number of reasons, one of which is that the characteristics
one wants to see in the names used by routing is rather at variance with the
characteristics one wants to see in the names used in end-end communications.
Noel
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
