[1]The graph's vertices represent either routers, transit networks, or stub networks. Since routers may belong to multiple areas, it is not possible to color the graph's vertices.
[2]It is possible for all of a router's interfaces to be unnumbered point-to-point links. In this case, an IP address must be assigned to the router. This address will then be advertised in the router's router links advertisement as a host route.
[3]Note that in these cases both interfaces, the non-virtual and the virtual, would have the same IP address.
[4]Note that no host route is generated for, and no IP packets can be addressed to, interfaces to unnumbered point-to-point networks. This is regardless of such an interface's state.
[5]It is instructive to see what happens when the Designated Router for the network crashes. Call the Designated Router for the network RT1, and the Backup Designated Router RT2. If Router RT1 crashes (or maybe its interface to the network dies), the other routers on the network will detect RT1's absence within RouterDeadInterval seconds. All routers may not detect this at precisely the same time; the routers that detect RT1's absence before RT2 does will, for a time, select RT2 to be both Designated Router and Backup Designated Router. When RT2 detects that RT1 is gone it will move itself to Designated Router. At this time, the remaining router having highest Router Priority will be selected as Backup Designated Router.
[6]On point-to-point networks, the lower level protocols indicate whether the neighbor is up and running. Likewise, existence of the neighbor on virtual links is indicated by the routing table calculation. However, in both these cases, the Hello Protocol is still used. This ensures that communication between the neighbors is bidirectional, and that each of the neighbors has a functioning routing protocol layer.
[7]When the identity of the Designated Router is changing, it may be quite common for a neighbor in this state to send the router a Database Description packet; this means that there is some momentary disagreement on the Designated Router's identity.
[8]Note that it is possible for a router to resynchronize any of its fully established adjacencies by setting the adjacency's state back to ExStart. This will cause the other end of the adjacency to process a SeqNumberMismatch event, and therefore to also go back to ExStart state.
[9]The address space of IP networks and the address space of OSPF Router IDs may overlap. That is, a network may have an IP address which is identical (when considered as a 32-bit number) to some router's Router ID.
[10]It is assumed that, for two different address ranges matching the destination, one range is more specific than the other. Non- contiguous subnet masks can be configured to violate this assumption. Such subnet mask configurations cannot be handled by the OSPF protocol.
[11]MaxAgeDiff is an architectural constant. It indicates the maximum dispersion of ages, in seconds, that can occur for a single link state instance as it is flooded throughout the routing domain. If two advertisements differ by more than this, they are assumed to be different instances of the same advertisement. This can occur when a router restarts and loses track of the advertisement's previous LS sequence number. See Section 13.4 for more details.
[12]When two advertisements have different LS checksums, they are assumed to be separate instances. This can occur when a router restarts, and loses track of the advertisement's previous LS sequence number. In the case where the two advertisements have the same LS sequence number, it is not possible to determine which link state is actually newer. If the wrong advertisement is accepted as newer, the originating router will originate another instance. See Section 13.4 for further details.
[13]There is one instance where a lookup must be done based on partial information. This is during the routing table calculation, when a network links advertisement must be found based solely on its Link State ID. The lookup in this case is still well defined, since no two network links advertisements can have the same Link State ID.
[14]This clause covers the case: Inter-area routes are not summarized to the backbone. This is because inter-area routes are always associated with the backbone area.
[15]This clause is only invoked when Area A is a Transit area supporting one or more virtual links. For example, in the area configuration of Figure 6, Router RT11 need only originate a single summary link having the (collapsed) destination N9-N11,H1 into its connected Transit area Area 2, since all of its other eligible routes have next hops belonging to Area 2 (and as such only need be advertised by other area border routers; in this case, Routers RT10 and RT7).
[16]By keeping more information in the routing table, it is possible for an implementation to recalculate the shortest path tree only for a single area. In fact, there are incremental algorithms that allow an implementation to recalculate only a portion of a single area's shortest path tree [BBN]. However, these algorithms are beyond the scope of this specification.
[17]This is how the Link state request list is emptied, which eventually causes the neighbor state to transition to Full. See Section 10.9 for more details.
[18]It should be a relatively rare occurrence for an advertisement's LS age to reach MaxAge in this fashion. Usually, the advertisement will be replaced by a more recent instance before it ages out.
[19]Only the TOS 0 routes are important here because all OSPF protocol packets are sent with TOS = 0. See Appendix A.
[20]It may be the case that paths to certain destinations do not vary based on TOS. For these destinations, the routing calculation need not be repeated for each TOS value. In addition, there need only be a single routing table entry for these destinations (instead of a separate entry for each TOS value).
[21]Strictly speaking, because of equal-cost multipath, the algorithm does not create a tree. We continue to use the "tree" terminology because that is what occurs most often in the existing literature.
[22]Note that the presence of any link back to V is sufficient; it need not be the matching half of the link under consideration from V to W. This is enough to ensure that, before data traffic flows between a pair of neighboring routers, their link state databases will be synchronized.
[23]When the forwarding address is non-zero, it should point to a router belonging to another Autonomous System. See Section 12.4.5 for more details.