3.4Comparing EIGRP and OSPF routing concepts

Being able to explain "why is this route in the routing table and that one is not" is the foundation of L3 troubleshooting. The decision happens in two stages: first, when multiple routing protocols offer the same destination, the lower administrative distance (AD) wins, and only then, within the same protocol, the lower metric wins. Reversing that order derails the entire diagnosis. This section compares two IGPs of very different character, EIGRP and OSPF, along four axes: algorithm, metric, load balancing, and hierarchical design.

3.4.1EIGRP: advanced distance vector and DUAL

• EIGRP descends from distance vector protocols that use distance information heard from neighbors, but because it maintains neighbor relationships, sends partial updates, and computes loop-free paths with DUAL (Diffusing Update Algorithm), it is called an advanced distance vector protocol. For each destination it keeps the FD (feasible distance), its own best metric to the destination, and the RD (reported distance), the metric a neighbor reported.
• The best path is the successor, and an alternative satisfying RD < FD (the feasibility condition) is a feasible successor, held in the topology table and promoted instantly without recomputation when the successor fails—hence very fast convergence. Only when no feasible successor exists does EIGRP send queries to neighbors (the active state), and if those spread, convergence slows (stuck-in-active).
• By default the metric is computed from bandwidth and delay (K1 and K3 active; reliability and load unused by default). EIGRP's unique strength is unequal-cost load balancing via variance: variance 2 lets traffic also use feasible-successor paths whose metric is within twice the FD, in proportion to their metrics. Its AD is 90 for internal routes, 170 for external, and 5 for summaries.

3.4.2OSPF: link-state and SPF

• In OSPF, every router floods link information (LSAs) within an area so that all routers hold an identical LSDB (link-state database), and each then runs SPF (Dijkstra) to build a shortest-path tree rooted at itself. Because routers share a map and compute for themselves rather than passing along hearsay, loops are structurally unlikely.
• The metric is cost, by default reference bandwidth 10^8 bps (100 Mbps) divided by interface bandwidth (truncated, minimum 1). Consequently every link of 1 Gbps or faster collapses to cost 1 and becomes indistinguishable, so in practice you raise it with something like auto-cost reference-bandwidth 100000 (a 100 Gbps reference) and must set the same value on every router in the area.
• Load balancing is equal-cost (ECMP) only; there is no unequal-cost feature equivalent to EIGRP variance. Its AD is 110. Hierarchy comes from areas, and every area must attach to the backbone (area 0). The boundary roles are the ABR (area border) and ASBR (external redistribution point).
• Choose the area type by requirement: normal (receives all LSAs), stub (blocks external type-5 LSAs, substituting a default route), totally stubby (Cisco-specific; also blocks type-3 inter-area routes, leaving essentially only a default), and NSSA (stub-like but permits an ASBR inside, injecting externals as type 7 which the ABR converts to type 5). All exist to shrink the LSDB and routing table.

3.4.3Axes for choosing

• In a multi-vendor environment, standard OSPF is effectively mandatory. On an all-Cisco network where you want to fully use several WAN circuits of differing bandwidth simultaneously, EIGRP wins thanks to unequal-cost load balancing. EIGRP is also simpler to configure, but at scale OSPF's explicit hierarchy (areas) gives better operational visibility.
• AD ranks protocols against each other, not metrics. If EIGRP (AD 90) and OSPF (AD 110) both offer the same destination, EIGRP is installed regardless of which metric looks better. In a dual-protocol network, the symptom "the shortest path OSPF computed is not being used" is usually the expected consequence of AD precedence, and changing it deliberately means adjusting distance.

Suppose you receive a report that WAN paths between a branch and headquarters are not being used as designed. There are two circuits toward headquarters, a 1 Gbps fiber link and a 200 Mbps backup, and the intent was "use the fast fiber primarily, and ideally use both in proportion to bandwidth." Told that OSPF is running, you check show ip route and find both circuits load-balancing at equal cost. That may look like "both are in use, so fine," but it means the 200 Mbps circuit carries as much traffic as the 1 Gbps one, congesting the slow link and degrading overall quality. The cause is OSPF cost calculation: with the default reference bandwidth of 10^8 bps (100 Mbps), 1 Gbps yields 10^8 / 10^9 = 0.1 and 200 Mbps yields 10^8 / (2 x 10^8) = 0.5, both of which are clamped to the minimum cost of 1, so the two are indistinguishable. There are two remedies. First, if you stay on OSPF, raise the reference bandwidth with auto-cost reference-bandwidth 100000 (identically on every router in the area) so the costs differ and the fast circuit wins. Note, however, that OSPF can only load-balance equal-cost paths, so this leaves only the fast circuit in use with the backup idle. Second, if you truly want the original requirement of using both in proportion to bandwidth, consider migrating to EIGRP—but only if the all-Cisco premise holds—because with variance 2, EIGRP sends traffic over feasible-successor paths whose metric is within twice the FD, in proportion to their metrics. A tempting partial solution to avoid is "manually equalize the OSPF costs so both are used." That merely produces equal-cost balancing, i.e. an even split, not a bandwidth-proportional one, and effectively cements the current problem. And if EIGRP and OSPF run simultaneously during the migration, you must account for the fact that AD (EIGRP 90 < OSPF 110) makes EIGRP routes win unconditionally, or you will hit the next round of confusion: "I adjusted the OSPF cost but nothing changed."

3.4.4Section summary

• EIGRP is advanced distance vector using DUAL, promoting a feasible successor (RD < FD) instantly; variance enables unequal-cost balancing and its AD is 90
• OSPF shares an LSDB and computes with SPF; cost = 10^8 / bandwidth collapses everything at 1 Gbps and above to cost 1, so raise the reference bandwidth uniformly on all routers. Balancing is equal-cost only and its AD is 110
• Route selection is two stages: AD, then metric. OSPF area types (stub, totally stubby, NSSA) exist to shrink the LSDB and routing table