The IP over ATM Mailing List Archive by date

[Hiroshi Esaki <hiroshi@ctr.columbia.edu>]

Hi Tim and Bryan, 

Thanks for the clarification.  
The things seem to be going clear to me. :-)  


  >  (1) Multi-address node, having more than one IP address with
  >      one physical interface, is allowed ?
  >       --> I guess this is allowed.
 tjs> I believe this to be true.  Again, I use a single Ethernet
 tjs> adapter assigned multiple IP addresses as my model.

Can we allow to use a single ATM address for multiple IP addresses
which belong to the same LIS, as well as the Ethernet case ? 
I wanted to allow this, if possible. 
Though Bryan said NO in his reply.. 
Some hack or tricky technique (e.g. gratuitous InARP) seems to exsist.  


  >  (2) In multi-address node, how InARP-reply is replied.
  >      (a) Can VC between ARP-server and client node be shared among
  >          the multiple IP addresses ?
  >          If this is Yes, multiple InARP-reply messages will be
  >          returned to the ARP-server, according to a single InARP-reque ?
 tjs> No, I believe that per RFC1293, only one or zero replies are to be
 tjs> returned for each request.  The reply should be from an address that
 tjs> sits on the same subnet.  So the ARP Server would send out one request
 tjs> for each LIS that it served.  Likewise, the multi-addressed client
 tjs> would do the same.  Of course, an implementation may choose to send
 tjs> gratuitous InArp replies just as some send gratuitous ARPs in addition
 tjs> to their compliance with the RFC.

Gratuitous InARP, that is exactly what I am thining to solve this 
problem, if multiple InARP messaging or packed InARP message is not 
allowed by RFC1577 (or by RFC1577+), in the case of VC sharing by 
multiple IP addresses. 
  Well, when we use gratuitous ARP, as well as gratuitous InARP, 
we will be able to solve the old and never updated ARP cache in 
the clients, according to host mobility.  
  If the use of gratuitous InARP/ARP is allowed, it seems to be 
very happy.  For example, we can multicast (broadcast into LIS) 
gratuitous ARP, when the mapping between IP address and ATM address 
is changed.  Then, every client nodes in the LIS can update the 
ARP cache to the latest information. 
  But, in this case, a single ATM address may have multiple IP addresses..


Now my question is whether a single ATM address can have multiple IP 
addresses, that belong to the same LIS ?  
 # whether to have a dedicated VCC for each IP address or to allow to 
   have a shared VCC for multiple IP addresses seems to be the different 
   issue from the above issue. 




  >     (b) Each IP address uses own ATM-VCC between the client node and
  >         ARP-server.
  >         Is this mandatry ? (related to question (a))
 tjs> I don't think this is mandatory.  Your implementation may choose to
 tjs> have multiple dedicated VCs, for management or some other reason,
 tjs> but at the cost of maintaining  duplicate channels.
OK, you say this is not mandatry, but is very reasonable way to do. 
When the VCC is shared by multiple IP addresses, some scheme (e.g. 
gratuitous InARP) must be used. 
 # I think this may not be considered in RFC1577. 

  bg> Yes an ARP server may support multiple LISs and use the same
  bg> ATM address to do so. If a client supports multiple LISs it
  bg> may choose to use one or multiple VCCs to the ARP server to
  bg> resolve requests.
OK, you mentioned in the case of multiple IP addresses that never   
belong to the same LIS.  
In this case, you want to allow that a single ATM address has 
multiple IP addresses (different LISes). 


  bg> The problem with (b) is that an InATMARP request to such a node
  bg> will only return one of the IP addresses, not the lot. 
  bg> The only way to support 
  bg> (b) would seem to be to extend InATMARP to require a chain
  bg> of replies, 
I think there is at least another way to do. 
As TJS pointed out, the use of gratuitous InARP seems to be 
possible. 


   >     (c) Each IP address has its own ATM address.  In this case,
   >         multi-address node has multiple ATM addresses.
   >         Is this mandatry for multi-address node ?
   >         --> I think it is not mandatry.
  bg> I believe that if a node has multiple IP addresses in the _same_
  bg> LIS then each must have a distinct ATM address, so I think I'm
  bg> disagreeing here.

"MUST have a dsitinct ATM address"... 
I do not like this requirement, though......
If having a distinct ATM address is "must", you have to perform 
ATM address registration whenever the edge node obtain a new 
IP address (e.g. attaching a new node in the Ethernet behind 
edge node)....   
Am I missing something ?  


If we allow to use gratuitous ARP message (this should be actually 
multicasted to all nodes in the LIS), the issue seems to be 
resolved.   
If we can allow this operation, "must have" can be replaced by 
"will have".  



Regards, 

Hiroshi Esaki