The IP Over NBMA (ION) Archive

[Albert Manfredi <manfredi@arl.bna.boeing.com>]

Diane Schuster wrote:
> 
> Hi,
> 
> It seems to me that the problem of using anycast to locate a server will
> apply to many more applications than just NHRP. Anycast, by definition,
> implies that any server that can provide this service is just as good as
> any other server that can provide the service.

Not really. Anycast implies that the topologically _closest_ server
answering to that particular anycast address is the best server to
use. In principle, that one topologically closest could also be the
_only_ one a particular host can use, at that time.

> If the ability of a
> server to serve a particular client depends on "routing realm"
> affiliation (which I would expect to be quite common), then the basic
> premise for using anycast is not met.
> 
> Perhaps a better service for NHRP and many other applications to use to
> locate an appropriate server would be "manycast." This would be a
> service provided by the NBMA network whereby there are MANY servers
> which can  provide the service being sought equally well. However, not
> ANY server in the NBMA network can provide the service equally well. The
> difference between manycast and group or multicast addresses is that
> only one server receives a packet sent to a manycast address (i.e. not
> all servers with the manycast address receive the packet).

Interesting idea. As far as I can tell, the real difference between
your manycast and anycast is that manycast would depend on a set of
criteria which are _not_ solely proximity, based on ATM topology.

Bert
manfredi@arl.bna.boeing.com