The Routing Over Large Clouds Mailing List Archive by date

[Grenville Armitage <gja@bellcore.com>]

My question on TTLs was intended to confirm my suspicion that
_both_ NHRP and CSRs affect the level of TTL decrementing
along a given IP path, and that behaviour in the face of IP loops
is the main area of difference between these technologies.

When functioning 'normally' a simple IP path such as:

	host - CSR -... - CSR - host
or
	host - NHS - ... NHS - host
	    \                 /
             --cut-through----

..decrements the TTL in forwarded packets identically in the CSR
and NHRP cases. (So traceroute and TTL based scoping are warped equally.)

However, the question I have for the CSR proponents is how well
does a CSR cope with IP loops? (Or, to paraphrase a private comment
from Joel, how do the CSR and NHRP models react when IP routing
screws up?)

We know NHRP detects loops during the query/resolution/response phase,
and so wont allow a VC to be created that would participate in a loop.

As far as I understood Masataka's reply, the CSR model relys
on the CSR breaking the internal short-cut as soon as IP level
routing algorithm's operation discovers the existence of an external
loop. After a brief transient of looping packets, the CSRs effectively
switch back to 'TTL decrementing' mode (aka normal router).

Is this right? Is it reasonable? (i.e. do IP routing protocols
detect loops 'fast enough' ? Will loops be rare enough that only
misconfigured research nets will suffer from them? Are these
the assumptions behind the CSR model?)

Reactions from the CSR crowd?

cheers,
gja
_________________________________________________________________________
Grenville Armitage           http://gump.bellcore.com:8000/~gja/home.html  
On the Internet, I AM a dog.