The Routing Over Large Clouds Mailing List Archive by date

[Grenville Armitage <gja@bellcore.com>]

There is a difference between NHRP's potential impact and that
of the CSRs/IPSwitches. CSRs optimise (remove) the IP processing
of packets passing along a concatenated VCC path that includes the CSR.
This is a good thing.

NHRP in sparse ATM cloud scenarios may well boil down to the same
thing.

However, consider NHRP in a scenario where the ATM cloud is
densely populated with switches, and the overlaid IP network is sparse
wrt routers. The ATM VCC established after the NHRP Reply may end
up being routed across the ATM cloud using a different set of
intervening switches than might have been traversed under hop-by-hop
conditions (note I said switches, not routers).

The CSR/IPSwitch model has IP packets traversing the same set of ATM
switching points as hop-by-hop IP routing would have it.

Whether this distinction is relevant in reality is not clear to me,
as it depends on the ratio of IP routers to ATM switches. An assumption
that switches outnumber routers seems to permeate some NHRP discussions
wrt CSRs, although I've rarely seen it explicitly stated. Does anyone
really think it will be true? Why?

cheers,
gja