EIGRP Explained: Features, Metrics, and When It Still Fits

EIGRP keeps showing up in networks that prize fast convergence and quiet control planes. It blends distance-vector simplicity with loop-free guarantees, so day-to-day operations stay predictable while failures resolve quickly. If you’re deep in Cisco ecosystems, EIGRP can still be the lowest-friction way to move routes without dragging policy into the IGP.

Modern topologies look messy—overlay VPNs, campus fabrics, cloud on-ramps, and plenty of mixed speeds. So we’ll stick to what’s durable: the packets EIGRP actually uses, how DUAL makes loop-free decisions, how the composite metric is computed (classic vs. wide), and where EIGRP still makes sense next to OSPF and BGP.

We’ll also cover pragmatic design moves—summarization, stubs, and variance—plus a clean checklist for troubleshooting neighbors and taming queries before they turn into stuck-in-active headaches.

Core EIGRP Features That Matter Today

EIGRP is an advanced distance-vector protocol that uses the Diffusing Update Algorithm (DUAL) to guarantee loop-free convergence while limiting churn. It runs directly over IP using protocol number 88 and implements its own selective reliability rather than relying on TCP/UDP. Control messages are multicast to 224.0.0.10 for IPv4 and ff02::a for IPv6 and unicast when needed.

Neighbor Discovery, Timers, and Defaults

Neighbors form per interface when parameters match. Default hello/hold timers are 5/15 seconds on high-speed links and 60/180 on low-speed NBMA, which balances failure detection with link stability. You can tune per interface, but measure loss and latency first; too-tight timers cause false downs and unnecessary reconvergence.

Packet Types and Reliability

EIGRP uses five packet types: hello, update, query, reply, and acknowledgement (ACK). Updates, queries, and replies are delivered reliably with sequence control; hellos are lightweight; ACKs confirm receipt where reliability is required. When a route goes “active,” SIA-Query and SIA-Reply keep neighbors from resetting while progress toward a solution is demonstrated.

IPv6 Behavior Worth Knowing

EIGRP for IPv6 mirrors IPv4 but uses link-local next hops and the ff02::a multicast group. Because neighbors rely on link-local addresses, adjacencies don’t require a shared global prefix on the link, which makes numbered and unnumbered designs straightforward.

Understanding EIGRP Metrics

EIGRP’s path choice hinges on a composite metric. Classic metrics use a 32-bit value, while wide metrics extend precision to 64 bits so very fast links aren’t collapsed into ties.

Classic Metrics (K1–K5)

By default, K1=1 (bandwidth) and K3=1 (delay) are active; K2 (load), K4, and K5 (reliability) are 0. With those defaults, the classic metric behaves like 256 × (10^7/minimum path bandwidth in Kbps + cumulative interface delay in tens of microseconds). In practice, you steer preference by setting accurate interface bandwidth and delay values; enabling load or reliability tends to create flapping and is rarely helpful.

Wide Metrics in Named Mode

Wide metrics introduce higher-precision throughput and latency terms, scale the metric to 64 bits, and add an optional K6 for extended attributes. Named mode exposes these knobs (along with rib-scale to keep costs numerically reasonable), but it doesn’t change K-values by itself. On platforms that support wide metrics, you gain better differentiation across 10/40/100G paths without deviating from the stable bandwidth+delay model.

Feasibility Condition, Successors, and Variance

For each destination, DUAL tracks a successor (best next hop) and any feasible successors (pre-validated backups). A neighbor qualifies as a feasible successor only if its reported distance (RD) is strictly less than your feasible distance (FD). That rule prevents loops. If a failure occurs and no feasible successor exists, the router issues bounded queries until a new loop-free path is found. Unequal-cost load balancing works via variance: eligible feasible successors whose metric is ≤ FD × variance are installed so you can share traffic across non-identical paths.

Design Tips That Punch Above Their Weight

These patterns keep EIGRP quiet and predictable as the domain grows.

Use Manual Summarization Aggressively

EIGRP supports per-interface summarization on arbitrary bit boundaries. Summaries shrink query scope, cap failure domains, and install local discard routes (distance 5) so transient losses don’t black-hole traffic. If you must leak an exception under a summary, use a leak-map.

Stub and Stub-Site for Hub-and-Spoke

Mark branch routers as stubs so they don’t receive queries they can’t answer. In overlays like DMVPN, stub-site (named mode) generalizes the idea to the location: hubs treat the site as a stub automatically, which reduces control-plane noise during partial outages. Stub-site and classic stub are mutually exclusive; use the one that fits your code train and topology.

Named Mode for New Builds

Named mode unifies IPv4 and IPv6 under one process with address families, exposes wide-metric and rib-scale controls, and gives you granular per-interface policy. It interoperates with classic configs, but for new deployments it’s easier to standardize on named.

Timers, Throttles, and Scaling

Default hello/hold values are a decent baseline. If you tighten them, confirm QoS guarantees for control traffic; drops during congestion create ghosts like reset neighbors and active states. Limit maximum-paths to what hashing can spread sensibly, and only add variance on circuits where feasible successors truly exist.

Path Preference Without Surprises

Prefer adjusting interface delay (or wide-metric latency) over bandwidth to shape path choice. Delay is additive and intuitive, while bandwidth interacts with shaping and policers. Tag redistributed routes to avoid loops when EIGRP touches OSPF or BGP. Summarize first, filter second.

Authentication and Integrity

Use key chains and modern digests (such as HMAC-SHA-256 where supported) to prevent rogue neighbors and accidental cross-joins. Rotate keys with lifetimes and keep the control plane policed; a small bit of hygiene prevents large, noisy outages.

Where EIGRP Still Fits Versus OSPF or BGP

EIGRP shines in Cisco-centric interiors that value quick, quiet convergence with minimal policy: campus cores/distribution, industrial networks, and overlay underlays where UCMP squeezes value from asymmetric uplinks. Its ability to summarize anywhere and install feasible backups makes brownfield growth less risky.

Pick OSPF When Interoperability or Standardization Wins

OSPF is the safe interior default for multi-vendor networks. Its area model, ubiquitous tooling, and predictable equal-cost multipath meet most design goals without vendor lock-in, even though it doesn’t do true unequal-cost balancing.

Pick BGP When Policy and Edge Scale Dominate

BGP carries the heavy policy: filtering, attributes, communities, and large table scale. Many designs run EIGRP or OSPF in access/distribution and summarize into BGP at the edge for clean separation of reachability and policy.

Quick Configuration Pointers

Start in named mode, define address-family ipv4/ipv6, set passive-interface default then enable only where needed, configure summary-address at boundaries, declare stubs (or stub-site), and—if you want UCMP—set a conservative variance (2–3) only where feasible successors exist and validate splits with telemetry.

Troubleshooting Shortlist

When neighbors won’t form, check matching AS/K-values, authentication, and L2 multicast reachability to 224.0.0.10 or ff02::a. When routes stick active, look for summarization gaps, missing stubs, or filtered return paths. If UCMP seems off, verify the feasibility condition; variance never installs non-feasible alternates.