Routing Metrics

We've explored individual routing metrics—hop count, bandwidth, delay, and reliability—each capturing a single dimension of path quality. But real network paths exhibit all these characteristics simultaneously. A path might offer high bandwidth with high delay, or low delay with questionable reliability. How do routing protocols reconcile these competing considerations?

The answer lies in composite metrics: mathematical formulas that combine multiple individual metrics into a single value for path comparison. Composite metrics represent the most sophisticated approach to routing metric design, enabling nuanced trade-offs between competing network characteristics.

Single-dimensional metrics inevitably make trade-offs that may not align with network requirements. Consider these scenarios where individual metrics fail:

The Multi-Objective Optimization Problem:

Path selection is inherently a multi-objective optimization problem. We want paths that are:

• High bandwidth (for throughput)
• Low delay (for responsiveness)
• High reliability (for consistency)
• Low cost (for economy)

These objectives often conflict—high-bandwidth submarine cables have higher delay than terrestrial paths; the most reliable paths may not be the fastest. Composite metrics provide a framework for making principled trade-offs.

The Generic Composite Metric Concept:

Composite Metric = f(Bandwidth, Delay, Reliability, Load, ...)

Where f() is a function that:
1. Normalizes different metrics to comparable scales
2. Applies weighting factors based on priorities
3. Produces a single comparable value for path selection

The specific function varies by protocol, but the concept remains consistent: transform multiple dimensions into one.

EIGRP (Enhanced Interior Gateway Routing Protocol) provides the most widely deployed composite metric implementation. Understanding EIGRP's formula reveals how composite metrics work in practice.

Why Bandwidth Uses Minimum (Bottleneck):

EIGRP uses the minimum bandwidth along the path rather than summing bandwidths. This reflects physical reality: the effective throughput of a path is limited by its slowest link.

Path: 10Gbps → 1Gbps → 10Gbps
Effective throughput: 1Gbps (the bottleneck)

EIGRP BW component = 10^7 / 1,000,000 = 10
(Not 10^7 / 21,000,000 which would be incorrect)

Why Delay Uses Sum (Cumulative):

Unlike bandwidth, delay adds up. Each link contributes its delay to the total.

Path: 10ms + 5ms + 10ms = 25ms total
EIGRP Delay = 25000 tens of microseconds / 10 = 2500

This asymmetry—minimum for bandwidth, sum for delay—accurately models how paths behave physically.

Let's work through complete EIGRP metric calculations to build intuition for how composite metrics behave in practice.

Key Insights from Examples:

1. Bandwidth has diminishing returns: The BW formula (10^7 / BW) means the difference between 1 Gbps and 10 Gbps is only 9 units, while the difference between 100 Mbps and 1 Gbps is 90 units. Improving already-fast links has less metric impact.

2. Delay scales linearly: Each microsecond of delay adds equally to the metric. High-delay links (satellite, intercontinental) accumulate significant metric penalties.

3. Both factors matter: Neither bandwidth nor delay dominates unconditionally. The metric provides balanced consideration of both.

EIGRP's K values allow administrators to customize how the composite metric weighs different factors. While changing from defaults is rarely recommended, understanding the capability illustrates composite metric flexibility.

Why K2 (Load) and K4/K5 (Reliability) Are Disabled:

As discussed in previous pages, dynamic metrics cause routing instability:

Load (K2=0 by default):

• Interface load fluctuates constantly
• Route changes based on load cause traffic shifts
• Traffic shifts change load patterns
• Creates oscillating feedback loops
• Transport protocols (TCP) handle congestion better than routing reroutes

Reliability (K4=0, K5=0 by default):

• Reliability measurements are noisy
• Transient errors trigger route changes
• Route changes during errors compound problems
• Architectural redundancy is more effective

Scenarios Where K-Value Tuning Might Be Considered:

While rare, some scenarios theoretically warrant K-value changes:

1. Delay-sensitive applications dominant: Voice/video traffic might benefit from higher K3 (delay weight)
2. Bandwidth-constrained environment: May want to emphasize bandwidth (though defaults usually work)
3. Mixed media networks: Distinguishing between high-delay satellite and low-delay terrestrial

Classic EIGRP metrics were designed for networks of the 1980s-1990s. As bandwidth increased dramatically, the classic formula encountered scaling limitations. EIGRP Wide Metrics (introduced with EIGRP Named Mode) address these issues.

The High-Speed Problem:

With classic metrics, high-speed links become indistinguishable:

BW calculation = 10^7 / BW_in_Kbps

10 Gbps:  10^7 / 10,000,000 = 1
40 Gbps:  10^7 / 40,000,000 = 0.25 → rounds to 1
100 Gbps: 10^7 / 100,000,000 = 0.1 → rounds to 1

All three calculate to the same BW component of 1!

Wide Metrics Solution:

Wide metrics scale the formula to provide differentiation:

Wide BW = 10^7 × 65536 / BW_in_Kbps

10 Gbps:  655,360,000 / 10,000,000 = 65,536
40 Gbps:  655,360,000 / 40,000,000 = 16,384
100 Gbps: 655,360,000 / 100,000,000 = 6,554

Now we have meaningful differentiation!

Composite metrics—or their components—can be strategically manipulated to influence traffic flow. This practice, known as metric-based traffic engineering, allows network engineers to implement routing policies without complex policy routing configurations.

With OSPF using simple bandwidth-derived costs and EIGRP offering sophisticated composite metrics, which approach is better? The answer depends on network requirements and operational capabilities.

Operational Reality:

In practice, the sophistication gap is smaller than it appears:

1. OSPF with manual tuning can achieve similar results to EIGRP's composite metric by manually setting costs to reflect delay where needed.

2. EIGRP with defaults uses only bandwidth and delay, which OSPF administrators can replicate through cost configuration.

3. Neither protocol typically uses dynamic load or reliability in production—the problematic features are disabled.

4. Modern overlay solutions (SD-WAN) increasingly handle application-aware routing, reducing the importance of IGP metric sophistication.

When Composite Metrics Provide Clear Value:

• Networks mixing satellite and terrestrial links (delay varies dramatically)
• Environments requiring unequal-cost load balancing (EIGRP variance)
• Large Cisco-only networks where EIGRP's features can be fully utilized
• Scenarios where multiple path characteristics genuinely differ and matter

We've completed our exploration of routing metrics with composite metrics—the sophisticated approach to combining multiple path characteristics into unified values for comparison. Let's consolidate our knowledge from this page and the entire module:

Module Conclusion:

Routing metrics form the decision-making foundation of network routing. From the primitive simplicity of hop count through the sophistication of composite metrics, each approach represents trade-offs between accuracy, stability, and operational complexity.

The key insights for network engineers:

1. Understand your metrics: Know exactly how your routing protocol calculates costs and what factors influence path selection.

2. Configure appropriately: OSPF reference bandwidth and EIGRP bandwidth/delay settings must reflect your actual network.

3. Prefer stability: Avoid dynamic metrics (load, reliability) in production; architectural redundancy provides more reliable outcomes.

4. Use metrics for engineering: Strategic metric manipulation enables traffic engineering without complex policy configurations.

5. Match complexity to need: Don't use sophisticated features you can't troubleshoot. Simple, understood configurations beat complex, mysterious ones.

With this foundation in routing metrics, you're prepared to understand how routing protocols like RIP, OSPF, EIGRP, and BGP use these metrics to build and maintain routing tables across networks of any scale.