The IP over ATM Mailing List Archive by date

Cell Relay Retreat>List Archive>month:1995-Jun> msg00037



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Multi-addressed hosts

  • From: tcolley@Baynetworks.COM (Tom Colley)
  • Date: Wed, 21 Jun 1995 15:17:55 -0400 (EDT)

> 
> Also another potential problem area is that some implementations expect
> an ARP server to do the _reverse_ lookup i.e. from ATM address to IP
> address, after a client receives a SETUP message. If this is a 1 to N
> mapping then the ARP server may pick the first IP address that matches
> the ATM address, whereas it may be another IP address that is sourcing
> the data. In practice I don't think there should be a problem here,
> but I don't think it is a safe assumption to make. We should probably
> also clarify the requirements for this reverse lookup. At the moment it
> is neither explicitly required or prohibited.
> 
> I'm curious to know the reason for having multi-addressed hosts on a LIS
> where the addresses are in the same subnet. (All InATMARP requests issued
> by an ARP server are relative to a particular subnet). Is it some form
> of redundancy or useful when transitioning to a new number scheme ?
> 
> Bryan Gleeson
> Adaptec.
> 

One case I can think of is a modem pool type of device: n serial lines on
one side, and one ATM UNI on the other.  Each serial line is represented by
a different IP address, which may be on the same subnet.  If each line is
represented by a separate ATM address, there is no problem.  If all of the
lines are represented by one ATM address, then the mapping from ATM address
to protocol address becomes one-to-many.  There is somewhat of a precedent
for this type of device: AppleTalk Remote Access works this way--multiple
protocol addresses are associated with a single MAC address.

For a device like this, the reverse lookup expecting a one-to-one
relationship is questionable.  When is a reverse lookup necessary?
I haven't given this much though, but the only thing that comes to mind is
for verification of IP address <-> ATM address relationships that the server
actually learns?

Strictly interpreting this excerpt from 1293 seems to preclude an InARP
response from each IP address on the same subnet of the requestor though:

  7.1.  Operation with Multi-Addressed Hosts

   In the context of this discussion, a Multi-Addressed host will refer
   to a host that has multiple protocol addresses assigned to a single
   interface.  If such a station receives an InARP request, it must
   choose one address with which to respond. To make such a selection,
   the receiving station must first look at the protocol address of the


Tom