Why does EIGRP use bandwidth + delay for its metric instead of hop count like RIP?
I just moved from the RIP chapter to EIGRP in the course and I'm a bit confused. RIP was simple — count the hops, lowest wins, max 15. But EIGRP apparently uses this composite metric with bandwidth and delay (and load/reliability?) and I don't really get why it's more complicated.
What problem does bandwidth + delay actually solve that hop count doesn't? And when EIGRP picks a path, which one "wins" — the fewest hops or something else? Trying to build the right mental model before I do the lab. 🙂
- Good question to ask early — the RIP-vs-EIGRP contrast is really a design lesson: hop count optimizes for fewest routers, metric optimizes for actual link quality. Different goals entirely. — ariel.lahav22,
2 Answers
RIP only counts hops, so it has no idea a link is fast or slow — a 3-hop path over Gigabit loses to a 1-hop path over a 64k serial link. EIGRP fixes that by building a composite metric from bandwidth and delay.
Default metric (K1=K3=1, K2=K4=K5=0):
metric = (10^7 / min_bandwidth_kbps + sum_of_delays / 10) * 256
- bandwidth = the slowest link along the path (the bottleneck), in kbps
- delay = the cumulative interface delay of every hop, in microseconds
- the
*256is just the wide-metric scaling
Two consequences worth internalizing:
- Bandwidth is a min, delay is a sum. Adding a fast hop barely moves the metric; adding a slow hop can dominate it.
- Only bandwidth+delay count by default. Load and reliability are in the formula but weighted 0 — leave the K-values alone unless you really mean it, and they must match on both neighbors or the adjacency won't form.
Lowest composite metric to a destination = the feasible distance, and that neighbor becomes the successor (installed route). A backup (feasible successor) is any neighbor whose advertised distance is less than your FD — that's the feasibility condition DUAL uses to guarantee a loop-free instant failover.
In the lab: build R1-R2-R3, run EIGRP, then show ip eigrp topology — you'll see FD and the reported distance side by side.
- The min-vs-sum point finally made it click for me — bandwidth is the bottleneck, delay accumulates. Thank you! Going to build the R1-R2-R3 topology now and watch the topology table. — 22maya2n2,
Ben nailed the formula; here's the practical angle. The number that trips people up is delay, because it's not measured — it's a static per-interface value you can read with show interfaces. So EIGRP's "cost" is deterministic, not live traffic-based (load is off by default for exactly that reason — you don't want routes flapping with congestion).
Quick way to see path selection in action:
show ip eigrp neighbors— confirm the adjacency is up first.show ip route eigrp— the installed successor.show ip eigrp topology— FD vs. the neighbors' advertised distance; that's where you spot a feasible successor.
If a path you expected isn't chosen, check the bottleneck bandwidth — one slow serial anywhere on the path caps the whole thing. And remember the bandwidth EIGRP uses comes from the bandwidth interface command, which is often wrong/default and doesn't change actual throughput — it only steers routing. That mismatch is the #1 "why did it pick that path?" gotcha.
- +1 on the `bandwidth` command caveat. Worth adding: on faster interfaces the classic 10^7 formula saturates, which is exactly why wide metrics (the *256 scaling) exist — otherwise a 10G and a 1G link would tie. — ben31833,