Deep dive into EIGRP's load-balancing and metric manipulation engine. Unequal-cost load balancing via variance, share-count ratios, offset-list metric inflation rules, inbound vs outbound propagation behavior, and maximum-path tuning for optimal bandwidth utilization.
Author: Hrishi Pradhan
Level: Basic to Advanced
Sections: 4 Theory
Diagrams: 4 Mermaid
Recommended: Complete Module B first
1. Unequal-Cost Load Balancing via Variance Multiplier
How EIGRP uses paths with different metrics simultaneously
Section 1
What, Where, How & Why
Variance allows EIGRP to install multiple paths to the same destination even when their metrics differ, unlike OSPF and RIP which require identical metrics for load balancing. Use it in enterprise networks to utilize all available bandwidth across unequal-cost WAN links. It solves the problem of idle backup links waiting for a primary failure.
%%{init: { 'theme': 'base', 'themeVariables': { 'fontSize': '12px' }}}%%
flowchart TD
Start["Successor installed FD = 158720"] --> Multi["Multiply FD by variance 158720 x 2 = 317440"]
Multi --> Check{"Is candidate metric ≤ 317440?"}
Check -->|Yes| Install["Install candidate as load-share path"]
Check -->|No| Reject["Candidate remains Alternate only"]
Install --> End1["Equal distribution based on share-count"]
Reject --> End1
style Start fill:#7C3AED,color:#FFF
style Multi fill:#F59E0B,color:#FFF
style Install fill:#10B981,color:#FFF
style Reject fill:#EF4444,color:#FFF
R4 also fails Feasibility (RD > FD) → never eligible regardless of variance
2. Share Count & Proportional Traffic Distribution
How EIGRP decides how much traffic each path receives
Section 2
What, Where, How & Why
Once multiple paths are installed, EIGRP does not split traffic 50/50. It uses a share-count ratio based on path metrics: lower-metric paths carry proportionally more packets. This solves the problem of slow links being overloaded and fast links being underused.
Result: Out of every 5 packets, 4 go via A and 1 via B
Practical Table
Path
Metric
Share Count
Traffic Ratio
Path A (primary)
158720
2
66%
Path B (variance)
317440
1
33%
Total
—
3
100%
Important: Share count is not the same as bandwidth. A 4:1 ratio does not mean a path is 4× faster — it means EIGRP sends 4 packets on one path for every 1 on the other. Actual throughput depends on link speed.
Per-route metric manipulation for path control and load balancing
Section 3
What, Where, How & Why
Offset-lists increase the metric of specific routes, allowing engineers to selectively make certain paths less preferred. Use them to force traffic through a specific link, enable per-route unequal-cost load balancing, or prevent a route from becoming a Feasible Successor. They solve the problem of tuning metrics without affecting all routes on an interface.
%%{init: { 'theme': 'base', 'themeVariables': { 'fontSize': '12px' }}}%%
flowchart TD
ACL["ACL matches route permit 10.1.10.0 0.0.0.255"] --> Offset["offset-list ACL in +1000 on Gi0/1"]
Offset --> InCheck{"Direction?"}
InCheck -->|Inbound| Direct["Add offset directly to local metric"]
InCheck -->|Outbound| Downstream["Neighbor sees offset ÷ 256 rounded up"]
Direct --> Topology["Topology table updated"]
Downstream --> Topology
Topology --> Result["Successor / FS recalculated"]
style ACL fill:#7C3AED,color:#FFF
style Offset fill:#F59E0B,color:#FFF
style Direct fill:#10B981,color:#FFF
style Downstream fill:#0EA5E9,color:#FFF
Core Rule: Offset-lists can only increase metrics, never decrease
ACL Purpose: Matches the route, NOT filters traffic
Inbound: Adds to the local metric directly (visible immediately)
Outbound: Affects downstream routers; offset is divided by 256 and rounded up
Force traffic to prefer a different path on this router only
Outbound (out)
When advertising a route to a neighbor
Downstream routers see the inflated metric
Influence path selection on neighbors without changing local behavior
Configuration Example
! Step 1: Match the specific route with a standard ACL
ip access-list standard MATCH_ROUTE
permit 10.1.10.0 0.0.0.255
! Step 2: Apply offset in the desired direction
router eigrp 1234
offset-list MATCH_ROUTE in 1000 GigabitEthernet0/1
! This increases the local metric for 10.1.10.0/24 via Gi0/1 by 1000
! Step 3: Verify the change
show ip eigrp topology | include 10.1.10.0
Production Warning
Offset-lists can accidentally destroy a Feasible Successor. If you inflate a backup path's metric so that its RD exceeds the FD, the path will fail the Feasibility Condition and enter Active state during a future failure. Always verify with show ip eigrp topology after applying an offset-list.
How many paths can EIGRP install, and when does it use them?
Section 4
What, Where, How & Why
The maximum-path command controls how many equal-cost paths EIGRP will install in the routing table. Older IOS defaults to 4 paths; IOS v15 and later support up to 32. This solves the problem of a low default limit hiding available bandwidth on dense mesh topologies.
Equal-Cost: Paths with identical metrics install automatically
Default max-path: 4 (IOS classic); 16–32 in modern IOS
Traffic Distribution: Round-robin per flow (CEF-based)
No Variance Needed: Only maximum-paths is relevant
Combined with Variance: Variance handles unequal costs; max-path handles quantity
Configuration
router eigrp 1234
maximum-paths 8
! Allows up to 8 equal-cost paths
! To combine with unequal-cost load balancing:
router eigrp 1234
maximum-paths 8
variance 2
! First, install up to 8 equal-cost paths
! Then, install any unequal-cost paths that satisfy the variance rule
Equal-Cost vs Unequal-Cost — When to Use Which
Scenario
Equal-Cost
Unequal-Cost (Variance)
Two links with same bandwidth
Automatic
Not needed
Primary 1 Gbps + Backup 100 Mbps
Not possible (metrics differ)
Use variance
Four equal links + one slower
Use maximum-paths
Combine both commands
FE to backbone with tiered WAN
Unusual — links typically differ
Standard practice
The Golden Rule of EIGRP Load Balancing
Variance handles unequal-cost paths, but the underlying path selection still respects the Feasibility Condition. A path that fails Feasibility cannot be installed regardless of variance. This means proper DUAL topology design (multiple feasible successors) is the foundation for any load-balancing strategy.