Interaction of NSIS related QoS with respect to local exsiting QoS reservations

Hi,

I am going through the documentation with respect to RSVP/COPS, and 
NSIS, and am confused on the expected behavior when an NSIS related 
entity communicates with an RMF on a local node. Consider a scenario 
wherein we have 4 nodes and 2 end hosts,

Host1 <--> N1 <---> N2 <---> N3 <---> N4 <---> Host2

Assume each node (N1 thru N4) have existing QoS policies that (a) 
classify packets, and (b) shape/police and implement some form of 
(W)RED. Also assume that the classification on these nodes is with 
respect to some combinations of DSCP values.

Consider now an NSIS-QoS entity on N1 initiates a signalling procedure 
to indicate the reservation of resources along the path N2, N3 and N4. 
Since the nodes N2, N3 and N4 already have local QoS policies, how would 
the requested resource reservations be handled by the RMFs on these 
nodes? For a flow F1 identified by node N1 for resource reservations 
along the path N2-N3-N4,
- it is possible that F1 is mapped to "class-default" on N3 and does not 
have any BW/delay guarantees
- it is also possible that F1 is mapped to a non-HP class on N2 and 
hence does not have any delay guarantees.
- Do these "signalled" resource reservations apply on both ingress and 
egress directions. In other words, are we looking at uni-directional 
resource reservations?

I am trying to visualize how end-to-end resource requirements are mapped 
onto existing per-node resource reservations. In addition, I am 
interested in knowing any existing literature on resolution of conflicts 
between the signalled-resource reservations and the existing per-node 
resource reservations.

Thanks,

Ravi Bhagavatula
San Jose, CA