Re: SMDS ARP

[[e-mail] (Joel Replogle)]

George,
  Here are some responses.  Take them for what they are worth.


> Another question is related to the type of internet ad-
> dresses which can be sent to the SMDS ARP.  As I understand
> it, ARP maps *host* addresses to transit network addresses.
> What if SMDS ARP maps *network* addresses as well?  In this way,
> a router attached to SMDS could learn the SMDS address of another
> router behind which are internet networks.

  This sounds either like bridging 2 SMDS subnets together, or like
proxy-ARP.  To me it makes sense to employ the layered model and
use a network-level protocol to go between subnets (in both the 
telco and the IP sense of "subnet").  ARP is used to map a network
address (IP) onto a MAC address (SMDS), and you never use the MAC 
address of an entity on a different subnet.
  How do you know what MAC address to use when sending to an entity
on a different subnet?  That's a network-level routing protocol issue,
and shouldn't be addressed on the data-link level.

> Another question: rather than broadcast ARP with the consequent
> cost of multicast, would it be possible to send ARP packets to
> a single "well-known" SMDS address?

  I'm assuming that the "well-known" SMDS address is really just some
kind of server and not some trick that goes beyond my simplistic 
understanding of SMDS.   Given this assumption, this certainly is
a potential solution.  What this requires, however is that the 
"well-known" server authoritatively know all the IP/SMDS address
mappings, which may not necessarily be true.  For example, if I 
change the IP number on a host on SMDS, I'd prefer that it be 
immediatly known to all the other SMDS clients, instead of needing
to notify the SMDS managers of the change, and thus waiting perhaps
minutes, or perhaps hours.
  This isn't such a big deal.  Right now I have one type of media on
which I use static tables for the MAC/IP mappings.  It works fine, but
if I had my druthers, I'd prefer it to be dynamic.  I think a more 
important point is to keep SMDS and IP as loosely coupled as possible;
it would be bad if the central office switch software had to change
everytime there was a change in IP or in routing protocols, for example.


Joel