I have been recently thinking about OSPF designs with discontiguous non-Area 0 areas. An example scenario is an organization is migrating to a new data center on R3 & R4, and for an extended period of time will have two data centers connected to the old core (R1 and R2) and the new core (R3 and R4).
The old core already connects to Area 6 supporting the old data center. A new section of Area 6 will connect to the new data center. But there are no links planned to interconnect the two Area 6 clouds. Different blocks of the ‘data center’ address space 10.6.0.0/16 will be allocated to each data center.
According to the RFCs, it is indeed legal to have multiple discontiguous non-zero areas with the same Area ID, and it is not necessary to implement a virtual link or GRE tunnel to setup them in a contiguous manner as it is with Area 0. This is because the area ID is only used to create a neighbor relationship. It is not in any routing updates. When an ABR builds LSAs for Area 0, it takes the routing information from each of its other areas and includes them into Type-3 Summary LSAs. Therefore, the other Area 0 routers do not know what areas Type-3 Summary LSA destinations belong to. The routers only know that in order to reach these destinations, the next hop is a given ABR.
The design scenario described can support area range summarization for each data center as long as a more specific summary supporting the networks for each data centers exists at closest set of ABRs. You do need to be very careful to not create a black hole with area range summaries. But I started thinking about whether or not it could be considered a good design.
I polled several of my colleagues. The use of discontiguous non-zero areas is not a common design for production networks.
One friend mentioned supporting a discontinguous non-Area 0 areas for a customer a transitional period. The customer was introducing OSPF into a section of their network and had planned out an area numbering scheme. However, there was a portion of the network they apparently did not know as well as they thought. The customer ended up with two areas using the same number, although it took them a while to discover it. The discontigouous non-Area 0 area worked fine. They kept the discontiguous area numbering during the migration period until they replaced an old Nortel switch that had the old area configured.
The consensus was that the issue with this design is not if it technically works, it is if the people supporting this design will understand it well enough to support it optimally. The normal assumption of most network personnel is that OSPF areas are intended to be unique and contiguous in a production environment. If the areas are not unique and contiguous, how would you know that this an intentional design, and not broken network? It is also confusing if you are trying to document the network – some ABRs have an interface in one area 6, some have an interface in another 6. Although you could refer to the areas as “old Area 6” and “new area 6”, this scenario does not map to normal assumptions of OSPF design.
My conclusion is that while discontiguous non-Area 0 areas can be supported as a short term measure, this is not a good long term design for production networks.