Back to EIGRP Guide Path selection and tuning

EIGRP K Values and Metric Formula: K1–K5 Explained

By default, EIGRP uses K1 and K3 to calculate a composite metric from minimum path bandwidth and cumulative delay; K2, K4 and K5 are disabled. Learn the formula, how DUAL compares the result and how to verify every input before tuning a path.

BandwidthDelayK-ValuesFD and RDVariance

Metrics at a Glance

An EIGRP metric is a path score. Lower is better. With default K-values, the score uses the slowest configured bandwidth in the path and the total configured delay across the path.

Default inputsBandwidth + delay
Best valueLowest metric
Default K-valuesK1=1 · K3=1
Neighbor ruleK-values must match
Do not confuse values: Bandwidth, delay, load, and reliability can affect the formula. MTU is carried in EIGRP information but is not a K-value metric component.

In This Lesson

  1. Compare two paths in one lab
  2. Build the metric mental model
  3. Calculate the bandwidth component
  4. Calculate the delay component
  5. Work through the classic formula
  6. Map K1 through K5 correctly
  7. Understand K-value neighbor mismatches
  8. Connect metrics to FD, RD, and DUAL
  9. Recognize wide metrics
  10. Use the bandwidth command safely
  11. Tune path choice with delay
  12. Use variance and unequal-cost paths
  13. Separate metric from administrative distance
  14. Verify every metric input
  15. Troubleshoot unexpected path selection
  16. Change metrics safely
  17. Complete guided metric labs
  18. Review the frequently asked questions

1. Compare Two Paths in One Lab

R1 can reach R4 through two paths. Do not choose by hop count. Calculate the minimum bandwidth and cumulative delay for each complete path.

Current successor

Path A: R1 → R2 → R4

  1. Minimum bandwidth: 100,000 Kbps
  2. Total delay: 1,000 microseconds
  3. Classic metric: 51,200
  4. Result: Lower score, so this path wins
Alternate path

Path B: R1 → R3 → R4

  1. Minimum bandwidth: 1,000,000 Kbps
  2. Total delay: 3,000 microseconds
  3. Classic metric: 79,360
  4. Result: Faster bandwidth but too much total delay
Lesson: A faster interface does not automatically produce the best route. EIGRP evaluates the complete composite metric.

2. Build the Metric Mental Model

Think of the calculation as four small steps. This is easier and safer than memorizing one long formula.

Step 1Find minimum bandwidth
Step 2Add all delays
Step 3Apply K-values
Step 4Choose lowest result
  • Bandwidth is a bottleneck value: Use only the lowest configured bandwidth along the path.
  • Delay is cumulative: Add the delay of every outbound interface used to reach the destination.
  • K-values are weights: They decide which components participate in the composite metric.
  • Lower wins: DUAL uses the metric while selecting the successor and testing alternate routes.

3. Calculate the Bandwidth Component

Classic EIGRP converts the slowest path bandwidth into an inverse score. A slower bottleneck creates a larger value, and a larger metric is less preferred.

Classic calculation

Bandwidth component =
10,000,000 / minimum bandwidth in Kbps

Path A:
10,000,000 / 100,000 = 100

Find the configured value

R1# show interfaces GigabitEthernet0/0
  MTU 1500 bytes, BW 100000 Kbit/sec,
  DLY 500 usec
Bandwidth is descriptive: The IOS bandwidth command does not change physical link speed. It changes the value that routing protocols and other features may use.

4. Calculate the Delay Component

Classic EIGRP adds path delay and converts microseconds into tens of microseconds. Every outbound interface on the path contributes.

Classic calculation

Delay component =
sum of delay in microseconds / 10

Path A:
(500 + 500) / 10 = 100

Compare a longer delay

Path B:
(1500 + 1500) / 10 = 300

More cumulative delay means
a larger EIGRP metric.

Delay is often the cleaner EIGRP tuning input because changing it does not misdescribe the physical capacity of the link. However, other features may also read interface delay, so check platform behavior before using it.

5. Work Through the Classic Formula

With default K-values, the full formula simplifies to bandwidth plus delay, multiplied by 256.

Default formula

Metric = 256 ×
(Bandwidth component + Delay component)

Path A:
256 × (100 + 100) = 51,200

Compare Path B

Bandwidth = 10,000,000 / 1,000,000 = 10
Delay = 3,000 / 10 = 300
Metric = 256 × (10 + 300)
Metric = 79,360
PathBandwidth partDelay partFinal metricResult
A10010051,200Successor
B1030079,360Alternate
Values vary by platform: Modern releases may display wide metrics. Use the output from your routers when comparing real paths; do not force a lab's sample number onto a different platform.

6. Map K1 Through K5 Correctly

K-values control the composite formula. The default design uses bandwidth and delay because they are stable configured properties.

K-valueMetric roleDefaultOperational note
K1Bandwidth1Enables the minimum-bandwidth component
K2Load term0Dynamic load can make path choice less stable
K3Delay1Enables cumulative interface delay
K4Reliability denominator0Used with K5 in the optional reliability factor
K5Reliability multiplier0When zero, the reliability factor is treated as one
MTUCarried route informationNot a K-valueNot used in the classic metric calculation
Default set: K1=1, K2=0, K3=1, K4=0, and K5=0. Leave these defaults unless a documented design has been tested across every device.

7. Understand K-Value Neighbor Mismatches

Neighbors must use compatible K-values because they need the same rules for comparing route metrics. A mismatch prevents the adjacency from forming.

1

Hello arrives

R1 receives an EIGRP Hello from R2.

2

K-values differ

The routers would calculate routes with different weights.

3

Neighbor rejected

No stable adjacency or route exchange is allowed.

Inspect the process

show ip protocols
show running-config | section router eigrp
show logging | include EIGRP

Example lab change

router eigrp 100
 metric weights 0 1 1 1 0 0

Do not use this example
in production without a design.

8. Connect Metrics to FD, RD, and DUAL

The metric is not only a route cost. DUAL also uses metric information to prove whether an alternate neighbor can provide a loop-free backup.

TermMeaningHow to use it
Feasible distance (FD)The local router's best known total metricRepresents the successor path's local distance
Reported distance (RD)The metric a neighbor reports to the destinationShows how close that neighbor says it is
Feasibility conditionAlternate RD is lower than the current successor FDProves the alternate does not loop back through this router
SuccessorBest loop-free next hopInstalled in the routing table
Feasible successorLoop-free alternate that passes the conditionKept ready for fast convergence
Important: A path can be loop-free but fail the conservative feasibility condition. In that case, it is not stored as a feasible successor and a failure may require Queries.

9. Recognize Wide Metrics

Classic metrics were designed when interfaces were much slower. Named-mode EIGRP can use wide metrics to represent modern high-speed links with more precision.

  • Larger metric range: Wide metrics reduce the chance that very fast links look identical after integer scaling.
  • Modern bandwidth: High-speed interfaces can be compared more precisely than with the classic bandwidth constant.
  • Platform output matters: Displayed topology and RIB metrics can use scaling that differs from the simple classic example.
  • Migration requires testing: Confirm metric compatibility, route choice, variance, and redistribution across software generations.

Use the classic formula to learn the logic. Use actual topology output to operate a real named-mode deployment.

10. Use the Bandwidth Command Safely

The interface bandwidth value is metadata. It can influence EIGRP and other control-plane features, but it does not rate-limit traffic or change negotiated Ethernet speed.

ActionWhat changesWhat does not change
bandwidth 100000The reported bandwidth becomes 100,000 KbpsThe physical interface does not become 100 Mbps
Correct a false valueRouting and feature calculations may improveNo extra physical capacity is created
Use bandwidth only for tuningEIGRP metric may changeOther bandwidth-aware features may also be affected unexpectedly
Preferred practice: Keep bandwidth truthful. If the goal is EIGRP path preference only, consider interface delay or an explicit policy after reviewing platform impact.

11. Tune Path Choice with Delay

Increasing configured delay raises the EIGRP metric for paths using that interface. Change it gradually and predict the new successor before applying it.

Before the change

show interfaces GigabitEthernet0/1
show ip eigrp topology 10.4.4.0/24
show ip route 10.4.4.0
show ip cef 10.4.4.4 detail

Controlled lab change

interface GigabitEthernet0/1
 delay 200

Verify the topology, route,
forwarding path, and backup.
  • Calculate both candidate paths before changing the value.
  • Confirm the intended route becomes successor and the alternate remains safe.
  • Test forward and return traffic because the other direction makes its own route choice.

12. Use Variance and Unequal-Cost Paths

variance allows qualified unequal-cost paths to enter the routing table. It does not make an unsafe path loop-free and does not automatically include every route under the multiplier.

TestQuestionRequired result
FeasibilityDoes the alternate RD satisfy the feasibility condition?Yes; the path must be a safe feasible successor
MultiplierIs the alternate metric within variance × successor FD?Yes; otherwise it remains outside the installed set
Maximum pathsIs there room for another next hop?Yes; maximum-paths limits installed paths

Lab controls

router eigrp 100
 maximum-paths 4
 variance 2

Prove the result

show ip eigrp topology all-links
show ip route 10.4.4.0
show ip cef 10.4.4.4 detail
Traffic warning: Multiple installed paths do not promise an exact traffic percentage. CEF hashing, flow patterns, and traffic-sharing configuration influence real distribution.

13. Separate Metric from Administrative Distance

EIGRP metric compares EIGRP paths. Administrative distance compares routes learned from different sources. Check the route source before troubleshooting the composite metric.

DecisionValue usedExample
Choose between static and EIGRPAdministrative distanceStatic AD 1 normally wins over internal EIGRP AD 90
Choose between two internal EIGRP pathsEIGRP composite metricThe lower bandwidth-and-delay result becomes successor
Identify redistributed EIGRP routeRoute code and ADD EX normally uses AD 170
Common mistake: Lowering an EIGRP metric will not beat a route from another source that has already won by administrative distance.

14. Verify Every Metric Input

Read the interfaces first, the topology table second, and the installed route last. This connects configured inputs to DUAL's decision.

Interface inputs

show interfaces
show ip eigrp interfaces detail
show running-config interface Gi0/0
show running-config interface Gi0/1

DUAL evidence

show ip eigrp topology 10.4.4.0/24
show ip eigrp topology all-links
show ip protocols
show ip route 10.4.4.0
CheckHealthy evidenceQuestion answered
Interface bandwidth and delayValues match the intended designWhat inputs are being advertised?
K-valuesDefaults or documented matching valuesWhich formula do neighbors use?
Topology entryExpected FD, RD, successor, and alternatesHow did DUAL compare paths?
Routing and CEF tablesExpected next hop or next hopsWhich path actually forwards traffic?

15. Troubleshoot Unexpected Path Selection

Start with the observed route source and metric inputs. Do not change K-values or variance until you can explain the current path mathematically.

SymptomLikely causeCheck next
Slower path becomes successorIncorrect bandwidth, accumulated delay, or route sourceBoth interfaces and the full topology entry
Expected backup is not feasibleReported distance fails the feasibility conditionRD versus current successor FD
Variance installs no extra pathAlternate is not feasible, outside multiplier, or maximum reachedAll-links topology and maximum paths
Neighbor disappears after metric changeK-value mismatchshow ip protocols on both ends
Displayed metric differs from hand calculationWide metrics, scaling, or different interface valuesSoftware mode, release, and actual output
Metric changed but route did notAnother route source wins by ADFull show ip route prefix output
  1. Record the chosen route source, next hop, AD, and metric.
  2. List every candidate path with minimum bandwidth and cumulative delay.
  3. Confirm K-values and whether the platform uses classic or wide metrics.
  4. Change only the confirmed wrong input, then repeat the same verification.

16. Change Metrics Safely

Metric tuning is a routing change. Define the expected successor, backup, traffic path, and rollback before applying it.

1

Baseline

Save interfaces, topology, routes, CEF, traffic, and CPU.

2

Predict

Calculate the intended metric and successor before the change.

3

Change one input

Adjust only the approved bandwidth, delay, or policy.

4

Prove or roll back

Verify both directions and restore if the result differs.

Watch convergence: A metric change can move traffic immediately. Check link capacity, application impact, and the return path—not only the routing table.

17. Guided Metric Labs

Use a four-router topology with two paths between R1 and R4. Restore the baseline before each task.

Task 1: Calculate Before Looking

  1. Write down bandwidth and delay for both paths.
  2. Find each minimum bandwidth.
  3. Add each path's delays.
  4. Calculate both classic metrics.
  5. Predict the successor, then verify it.

Expected result: Your manual comparison agrees with the topology table.

Task 2: Tune with Delay

  1. Capture the current successor and feasible successor.
  2. Increase delay on one lab interface.
  3. Predict whether the successor will change.
  4. Verify topology, route, CEF, and traffic.
  5. Restore the original delay.

Expected result: You influence path choice without falsifying link bandwidth.

Task 3: Explain a Missing Backup

  1. Display all links for one destination.
  2. Record successor FD and alternate RD.
  3. Test the feasibility condition.
  4. Explain why the alternate is or is not feasible.
  5. Fail the primary only in the isolated lab and observe DUAL.

Expected result: You connect metric values to immediate failover or a Query process.

Task 4: Test Variance Carefully

  1. Confirm the alternate is a feasible successor.
  2. Calculate the smallest useful variance.
  3. Configure variance and maximum paths.
  4. Inspect routing and CEF next hops.
  5. Remove the change and confirm the baseline.

Expected result: You know why a multiplier alone cannot install an unsafe path.

18. Frequently Asked Questions

Which values does EIGRP use by default?

Default EIGRP uses minimum path bandwidth and cumulative path delay. The default K-values enable K1 and K3.

Is MTU part of the EIGRP composite metric?

No. MTU can be carried with EIGRP route information, but it is not a K-value component of the classic metric formula.

Does the bandwidth command change interface speed?

No. It changes descriptive interface metadata used by EIGRP and possibly other features. Physical speed and traffic shaping are controlled separately.

Why must K-values match?

Neighbors need the same metric rules. If their K-values differ, they cannot safely compare routes and the adjacency is rejected.

Should I enable load and reliability in the metric?

Usually not. They can change dynamically and make route selection harder to predict. Most designs keep default K-values and tune stable configured properties or policy.

Why did variance fail to install an alternate path?

The path may have failed the feasibility condition, exceeded the variance multiplier, or reached the maximum-paths limit. Check the all-links topology output.

Why does my displayed metric differ from the classic formula?

Your platform may use wide metrics or scaling, or the real interface values may differ from the example. Compare software mode, interface bandwidth, delay, and topology output.