> [Lucy] I see this as shifting the complexity to operator, which is > worse than what I propose.
Not sure why that would be... Operators will surely have the capability to provision multi-VN services and the allowed interconnections of VNs. This will be needed in any case. So one could also argue it would be simpler to leverage the same functionality to handle the multi-subnet case. And if you want VNs to know about subnets (so that each subnet is its own broadcast domain), now the VN needs to know what subnets actually are and who (which TS's) belongs to what subnet. That would seem to add complexity to the L3 VN case... > I agree that this is more complex service than pure L2 overlay or > pure L3 overlay. If we define it as a new service, we can have a > solution for the operator. This will avoid an operator to construct > individual L2 overlays for the subnets and an L3 overlay to > interconnect them for a tenant virtual network. Whether this is a "new service" or just a variation of an existing services seems less important than understanding what service needs to be provided to a tenant and how best to do that. Thomas _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
