OSPF models networks as
– Broadcast links
– Point to Point (P2P)
– Point to Multi-Point (P2MP)
– Non-Broadcast multi-access Networks (NBMA)
IS-IS models networks as
– Unnumbered Broadcast
The key differences are the way OSPF provides support for NBMA networks and inherent protocol support for unnumbered broadcast by IS-IS
(i) Support for NBMA Networks
IS-IS has no direct support for connecting ISs over a NBMA network and it must be modeled as a LAN or treated as a set of P2P links. Modeling it as the latter involves a lot of configuration and if full connectivity is not configured, multiple hops might be required for traversing the NBMA cloud.
Experience with ATM LAN emulation has proven un-scalable and insufficiently reliable because of the single point where replication takes place to emulate multicast.
The best alternative for IS-IS is thus to treat each PVC as a point-to-point link. All PVC failures are handled by the protocol since each PVC is visible to the protocol.
IS-IS mesh groups may be used to address the scaling issues which may result from redundant flooding in the highly meshed environments.
In OSPF there is a “NBMA mode” in the original specification which makes the protocol aware that it is on a NBMA network.
Neighbours are discovered initially through configuration which is restricted to the ones eligible for the DR election. To make administration easier and to reduce the HELLO traffic, most of the other routers attached to the NBMA subnet are assigned a router priority of zero. It thus involves quite a bit of administration overhead and is prone to mis-configuration. Also the network will malfunction if one of the nodes loses its link to the DR.
In this mode, each node in the NBMA must have a PVC to the DR and BDR. Since adjacencies between non-DR nodes is not mandated, the order of the number of adjacencies is O(2n), rather than O(n^2) as required when running OSPF without NBMA mode.
NBMA networks are thus only as robust and reliable as the underlying data-link service. If for example, a PVC fails or is mis-configured or if an SVC cannot be established, due to capacity or policy reasons, routing over NBMA subnet will fail. And, unfortunately, often the reason for the failure will not be immediately obvious to the network operator.
The P2MP can be applied to rectify these problems, although at some loss of efficiency.
(ii) Point-to-Multipoint model
This model can be used on any data link technology that the NBMA model can be used on. In addition, the P2MP model doesn’t require all the participating routers to be able to communicate directly to model a partial PVC mesh as a single P2MP networks. Dropping the full mesh requirement also allows the modeling of more exotic data link technologies, such as packet radio, as P2MP networks.
So if your favorite OS can’t support virtual interfaces or if there’s too much overhead involved in generating separate sub interfaces to each of the 500 ATM circuits then P2MP is good and can be your life saver!
However, when operating a full mesh Frame Relay or ATM network in P2MP mode, the work involved in neighbor maintenance, flooding, and database representation increases as O(n^2), where n is the number of OSPF routers attached to the subnet, instead of O(n)behavior that can be achieved with the original NBMA model.
(iii) Unnumbered broadcast
IS-IS supports unnumbered broadcast interfaces; however, most implementations do not! The protocol provides all necessary routing information without the aid of ARP, but doing this requires that each FIB entry contain a next-hop (circuit, SNAP address) pair for each path to a destination, and many routers are designed with FIB entries that contain only next-hop IP addresses instead, to reduce the size of the FIB and perhaps as a simplification.
For this reason, many implementations won’t interoperate with an unnumbered broadcast interface, and won’t interoperate with an implementation that doesn’t support ARP.