The IP Over NBMA (ION) Archive

["Andre N. Fredette" <fredette@BayNetworks.com>]

After receiving a couple of notes, I feel I should make my comments more
clear.

At 05:15 PM 8/19/97 -0400, Andre N. Fredette wrote:
>The OSPF Address Resolution Advertisement (ARA) Option proposed by Rob
>Coltun and Juha Heinanen seem to be very useful for ATM (or NBMA) networks.
> They allow OSPF to be used to distribute shortcut addresses as an
>alternative to using NHRP.

I believe ARA is an alternative to NHRP in certain environments.  For 
example, within an OSPF area and between ATM-attached routers ARA can 
replace NHRP.  

NHRP is needed when there are host NHCs and in the MPOA/LANE environment 
where subnets are distributed across multiple edge devices with different 
ATM addresses.  To use ARA in these cases, one would need to advertise host 
routes which is probably not desirable.  

A query mechanism also useful in cases when address information is 
aggregated by the routing protocol such as between OSPF areas, or routing 
domains (though it is arguable whether NHRP should be used in these cases 
either).

I believe that ARA complements NHRP quite nicely.

>
>Some people have proposed performing similar shortcut routing on broadcast
>LANs (e.g., Ethernet).  Given multiple subnets on a single broadcast LAN
>(perhaps using 802.1Q VLANs) NHRP or ARA could be used to obtain MAC
>addresses for inter-subnet traffic.
>
>This approach requires a single spanning tree to be maintained for the
>entire switched ethernet domain over which the shortcuts are desired.  It
>is my belief that this application of shortcuts doesn't scale well and is
>of very limited utility.
>

I do not want to get into an NHRP scalability argument here.  We've all 
experienced enough of that :-).  My comment was with respect to the scaling 
of ATM switching vs. Ethernet switching.  ATM addresses are similar to IP 
addresses in the respect that they can be summarized and aggregated, and a 
link-state routing protocol (PNNI) can be used to find the location of an 
address.  Ethernet addresses do not have these properties; therefore, 
spanning tree and bridge learning are used to help figure out how to deliver 
frames on Ethernet switches.  Current technology does not allow a switched 
connection to be set up across a number of ethernet switches in the way a VC 
can be established across a large ATM network.

Because of these properties, I make two conclusions:
1. It is feasible to build fairly large switched ATM networks over which 
shortcuts might be useful.
2. The size of Ethernet switched networks should be kept smaller, and IP 
routers and switches should be used for delivering traffic between them.

>Andre
>
>At 03:43 PM 8/19/97 -0400, Albert Manfredi wrote:
>>George Swallow's minutes included mention of OSPF advertisements which
>>incorporate layer 2 addresses, if I understood correctly. Did I get that
>>right?
>>
>>For ATM nets, maybe these could be used to create shortcuts. For IEEE
>>802 nets, what good are the layer 2 resolutions? I mean, is it an
>>extension to the proxy ARP concept, to include other networks? Will the
>>routers keep some sort of layer 2 table, for faster routing? Perhaps a
>>very quickly-aged table, of restricted size?
>>
>>If yes, it would be strange to see how an internet of IEEE LANs is
>>becoming more similar to ATM (shortcuts, switch-based, apparently
>>optimized for unicast rather than multicast).
>>
>>Can anyone clue me in? Thanks.
>>
>>Bert
>>manfredi@arl.bna.boeing.com
>>
>