At 04:24 PM 11/24/2004, Curtis Villamizar wrote:
> I don't see why that would be true. Remember that in that case of the
> tunnels appraoch, the whole essense of the approach is that the paths are
> unaffected by the failures. i.e. Fspace (aka Pspace) is the set of nodes
> which can be reached from S by paths which are unaffected by the failure,
> and Gspace (aka Qspace) is the set of nodes which can reach the "target"
> (proxy for destination) by paths which are unaffacted by the failure.
> the entire path tunneled to F/G and released to the destination is
> guaranteed to be unaffected by the failure.
The first part of your tunneling (since it isn't explicit route
tunneling) relies on IP forwarding. Plain IP with a multiple link
failure can have loops form even with symetric routes due to some
nodes reacting to the failure earlier than others.
I think that Mike (and I) are using an implicit assumption. There are two
potential causes for loops. The first case is when the use of the
alternates themselves cause loops. This could occur if, as described in
the LFA draft, two routers both have link-protecting LFAs that point at
each other and the node fails, so that both those routers invoke their
alternates. I believe that Mike was arguing that an alternate could be
found which prepares for an SRLG failure, and therefore one wouldn't have
that type of loops.
The second potential cause for loops would be due to the different new FIB
installation times of routers in the network. That is being addressed
separately in the micro-loop prevention work. Take a look at
http://www.psg.com/~zinin/ietf/rtgwg/IETF61-rtgwg-uloop-prev.ppt for the
presentation that Mike, Stewart and I did at the last IETF.
Rtgwg mailing list