On May 5, 2010, at 1:20 AM, Inder Monga wrote:
JOhn,
Thanks for putting thought into error management as well. I have a few questions though -
1. I don't think maintaining state is hard - every point to point protocol has state on either side whether it be PPP, IPsec/IKE, TCP etc. This is not just ATM.
The problem with state here is that it has to be maintained for all connection reservations, but all NSAs involved with making a reservation. The set of NSAs involved in a reservation for a particular connnection may be different depending on who made the reservation. If an NSA goes down the state of the connection as a whole becomes indeterminate. Or so it seems to me.
2. Each NSA pair maintains state between them. What they do is to inform the service tree under them when an error happens that requires notification up and down the service tree. The biggest issue here is a "state cleanup" problem i.e. when things go bad, you need to ensure that every NSA will clean up its state.
3. On slide 3, I do not understand the requirement "
•Signals from NRM and Requestor should be the same
–Need to define what these are, because they are carried on NSU
By this I mean that what the NRM signals about local resources should be the same as what gets signaled by NSI about remote resources. I think we agreed to this on the calls.