EIGRP packets, neighbour adjacency, timers and the neighbour and topology tables

EIGRP packets and reliable transport

EIGRP runs directly over IP as protocol 88. Like OSPF it has no TCP underneath, so it brings its own Reliable Transport Protocol (RTP). Important packets carry a sequence number and must be acknowledged; routine ones do not.

Hellos and most updates go to the multicast address 224.0.0.10. Replies, acknowledgements and retransmissions are sent as unicast. If a reliable packet is not acknowledged within the retransmission timeout, it is resent to that neighbour individually.

What must match to become neighbours

• Autonomous system number (classic mode) must be the same on both routers.
• K values (metric weights) must be identical.
• The interfaces must be on the same subnet, using their primary addresses.
• Authentication, if used, must match.
• The interface must not be passive, and no ACL may block protocol 88.

Unlike OSPF, the hello and hold timers do not have to match. Each router tells its neighbour how long to wait (its hold time) inside its hellos.

Hello and hold timers

By default EIGRP sends hellos every 5 seconds with a hold time of 15 seconds. On slow multipoint WAN interfaces (1.544 Mbit/s or less) the defaults are 60 and 180 seconds. If no EIGRP packet arrives from a neighbour before its hold time runs out, the neighbour is removed and its routes are recalculated. Because the hold time is advertised to the neighbour, changing it on R1 changes how long R2 waits for R1.

interface GigabitEthernet0/1
 ip hello-interval eigrp 100 2
 ip hold-time eigrp 100 6

Changing the hello interval does not change the hold time automatically; set both. For really fast failure detection, BFD is again the better tool.

Passive interfaces, static neighbours and the router ID

Passive interfaces work as in OSPF: the subnet is advertised, but no hellos are sent or accepted, so no neighbour can form. Use passive-interface default and enable only router-facing links.

A static neighbour makes EIGRP talk to one neighbour by unicast instead of multicast. It is useful over links that do not carry multicast well. Configuring it stops EIGRP processing multicast packets on that interface, so the neighbour must be configured statically on both sides.

router eigrp 100
 eigrp router-id 10.0.0.1
 network 10.0.0.0 0.0.255.255
 passive-interface default
 no passive-interface GigabitEthernet0/1
 neighbor 10.0.12.2 GigabitEthernet0/1

The EIGRP router ID is chosen like OSPF's (configured, then highest loopback, then highest interface address). It does not have to be unique to form an adjacency, but it is used to stop external routes looping: a router ignores an external route tagged with its own router ID. Duplicate router IDs therefore cause redistributed routes to disappear. Set it by hand.

Reading the tables when troubleshooting

show ip eigrp neighbors
show ip eigrp topology
show ip eigrp topology all-links
show ip eigrp interfaces detail
show ip protocols

In the neighbour table, Hold should count down from 15 and jump back up; Uptime that keeps resetting shows a flapping neighbour; a Q Cnt that stays above zero means reliable packets are queued and not being acknowledged, often a one-way link problem. In the topology table, P means a route is passive (stable) and A means it is active, with queries outstanding. Each entry shows the feasible distance and, per neighbour, the computed and reported distance.

Real-world example: at a logistics company, a new router at a depot would not peer with the regional hub. show ip protocols on each showed the same networks, but the depot router's log showed a K-value mismatch. A template copied from a test lab had set metric weights 0 1 1 1 0 0 to experiment with load. Restoring the default weights brought the adjacency up at once, and the team added a check for K values to its build review.