EIGRP vs OSPF — Differences, Metrics, Convergence and Which to Choose

Both are interior gateway protocols that route inside a single organisation, and either will run a mid-size enterprise network well. The real differences come down to three things: the algorithm, how failover works, and whether you are locked to one vendor.

Full Comparison

The Algorithms, Concretely

OSPF, everyone holds the same map

Each router floods LSAs describing its own links. Every router in the area assembles an identical link-state database, then runs Dijkstra against it to build a shortest-path tree rooted on itself.

The strength is completeness, the router knows the topology, not just what a neighbour claims. The cost is that any topology change means every router in the area reruns SPF. That is why areas exist: to keep the blast radius of a flapping link small.

EIGRP, DUAL and the feasible successor

EIGRP does not build a map. It keeps a topology table of what neighbours advertise, and for each destination it tracks:

• Successor, the best next hop, installed in the routing table.
• Feasible successor, a backup next hop that is guaranteed loop-free, precomputed and held in reserve.

A path qualifies as a feasible successor if its reported distance (the neighbour’s own distance to the destination) is less than the current feasible distance. That condition guarantees the neighbour is not routing through you, so switching to it cannot create a loop.

When the successor fails and a feasible successor exists, EIGRP installs it immediately, no recalculation, no query, sub-second. If no feasible successor exists, the route goes active and EIGRP queries its neighbours, which is much slower and is where EIGRP’s failure mode lives.

Router# show ip eigrp topology
Router# show ip eigrp topology all-links

The stuck-in-active problem

If a query goes unanswered, a dead neighbour, a congested link, a flat network with too many peers, the route stays active until the SIA timer expires and EIGRP tears down the adjacency. On a large flat EIGRP network, queries propagate far and this becomes a real operational problem.

The fix is bounding the query domain with summarisation and stub routers:

Router(config-router)# eigrp stub connected summary
Router(config-if)# ip summary-address eigrp 100 10.20.0.0 255.255.0.0

Marking spoke routers as stubs means they are never queried at all, which is the single most effective EIGRP scaling measure.

Metrics

OSPF cost

cost = reference bandwidth ÷ interface bandwidth

The 100 Mbps default reference means gigabit, 10-gig and 100-gig links all score a cost of 1 and are treated as identical. Raise it, and do so consistently on every router:

Router(config-router)# auto-cost reference-bandwidth 100000

EIGRP composite metric

By default EIGRP uses the slowest bandwidth along the path and the cumulative delay. The K-values that weight the components are K1=1, K3=1, and the rest zero, so load and reliability are not used, and enabling them is strongly discouraged because they change constantly and cause the topology to churn.

A practical consequence: because EIGRP includes cumulative delay, adjusting delay on an interface is the correct way to influence EIGRP path selection. Changing bandwidth affects other things too, QoS reservations, SNMP reporting, and is the wrong lever.

Router(config-if)# delay 1000

Unequal-Cost Load Balancing

This is genuinely unique to EIGRP. OSPF only load balances across equal-cost paths. EIGRP can use a slower backup path proportionally:

Router(config-router)# variance 2

With variance 2, any feasible successor whose metric is up to twice the successor’s is also installed, and traffic is distributed in inverse proportion to metric. Where you have a fast primary and a slower secondary link that would otherwise sit idle, this uses both.

Which Should You Choose?

Honestly: for most enterprises the choice is decided by vendor mix and existing skills, not by protocol capability. Both will route a few thousand prefixes without difficulty. OSPF is the safer default because it does not constrain future hardware choices.

Running Both

Networks end up running both after mergers or during migrations. Redistribution between them needs care:

• Redistributed routes enter EIGRP as external, AD 170, worse than OSPF’s 110, which changes path selection in ways people do not anticipate.
• Mutual redistribution at more than one point creates routing loops unless you filter with route maps and tags.
• Tag routes on the way out and deny tagged routes on the way back in. This is the standard safeguard and it is not optional.

Router(config)# route-map OSPF-TO-EIGRP deny 10
Router(config-route-map)# match tag 110
Router(config)# route-map OSPF-TO-EIGRP permit 20
Router(config-route-map)# set tag 100

Frequently Asked Questions

Which converges faster, EIGRP or OSPF?

EIGRP, when a feasible successor exists, the switch is immediate. Without one, EIGRP must query its neighbours and can be slower than OSPF. OSPF’s convergence is more consistent, if not the fastest case.

Is EIGRP proprietary?

Cisco published EIGRP as informational RFC 7868 in 2016, but adoption by other vendors has been minimal. In practical terms, running EIGRP means running Cisco.

Can EIGRP and OSPF run on the same router?

Yes. Both processes run independently and routes are chosen by administrative distance, EIGRP’s 90 beats OSPF’s 110 for identical prefixes.

Why does EIGRP have two administrative distances?

Internal routes (90) were learned natively by EIGRP. External routes (170) were redistributed in from elsewhere, and are trusted less because their origin is outside EIGRP’s own view.

Does OSPF support unequal-cost load balancing?

No. It load balances only across equal-cost paths. EIGRP’s variance has no OSPF equivalent.

What is a feasible successor?

A precomputed backup route that is guaranteed loop-free because the neighbour’s own distance to the destination is lower than yours. It can be installed instantly when the primary fails, which is the source of EIGRP’s fast convergence.

Which is better for a WAN?

EIGRP suits hub-and-spoke WANs well, stub routers, easy summarisation anywhere, and unequal-cost balancing across links of different speeds. OSPF works too, but its area structure fits a campus more naturally than a spoke topology. More context in the routing protocols guide.