The Routing Over Large Clouds Mailing List Archive by date

[Jon Crowcroft <J.Crowcroft@cs.ucl.ac.uk>]

 >> Now, why should there even be a transient loop of flow?
 >Now? Read the mail you are responding:
 >	a bug in the routing code can result in infinitely looping packets.
 
hence the expression....
ATM: past its cell by date

.....couldn't resist......:-)

but seriously, at the ATM cell forwarding layer, we don;t need ttl
type mechanisms to do failsafe loop breaking, since there's a VCI, and
if a VCI lookup finds an entry that points at te switch port a cell
arrived by, we ring alarms....

at the IP level, in the ipsilon flow labelling protocol, it should be
possible to preserve this loop freeness (i don't know if they do
though)

by the way, you don't need _bugs_ in datagram routing algorithms 
to get loops, you just need transients in distance vector schemes....


IPv4 TTL has two semantics:
1/ router hop (and whether a CSR counts as a router hop once it has
labelled a flow and is in cell stream/cut through mode, i don';t know
- i doubt it if it doesnt recalaculate routes...)
2/ time bounding the lifetime of packets to ensure TCP correctness
by decerementaing TTL per second of queueing time

the latter, in a very slow CSR ATM net, might be important (e.g. very
busy net with a slow ABR service.....) but i hope not:--)

 jon