If hop count represents the most primitive measure of network distance, then bandwidth represents the leap toward understanding actual network performance. Bandwidth—the theoretical maximum data transfer capacity of a network link—fundamentally transformed how routing protocols evaluate paths.

In the previous page, we saw how hop count could route traffic over a 56 Kbps link when a 10 Gbps alternative existed just one hop further away. Bandwidth-based metrics solve this problem by incorporating link capacity into routing decisions, ensuring that traffic flows through the fastest available paths rather than the shortest.

Bandwidth in networking refers to the maximum rate at which data can be transferred over a network link, typically measured in bits per second (bps) or its multiples (Kbps, Mbps, Gbps, Tbps).

Formal Definition:

As a routing metric, bandwidth represents the data-carrying capacity of a network link. Routing protocols use bandwidth values to calculate path costs, with higher bandwidth links receiving lower costs (making them more attractive for routing).

The Inverse Relationship:

Unlike hop count where lower is always better, bandwidth metrics operate inversely:

• Higher bandwidth = Lower cost = More attractive path
• Lower bandwidth = Higher cost = Less attractive path

This inversion requires protocols to convert bandwidth into a cost value. The most common conversion formula divides a reference bandwidth by the interface's configured bandwidth:

Critical Insight: The Minimum Cost Problem

Notice that with OSPF's default 100 Mbps reference bandwidth, all interfaces faster than 100 Mbps calculate to the same cost of 1. This means:

• 100 Mbps FastEthernet: Cost 1
• 1 Gbps GigabitEthernet: Cost 1
• 10 Gbps TenGigE: Cost 1
• 100 Gbps ports: Cost 1

OSPF treats a 100 Mbps link identically to a 100 Gbps link! This default behavior defeats the purpose of bandwidth-based routing in modern high-speed networks.

The reference bandwidth (also called auto-cost reference bandwidth) is the baseline value used to calculate interface costs. It must be configured consistently across all routers in an OSPF area to ensure consistent routing decisions.

The Consistency Requirement:

Every router within an OSPF area MUST use the same reference bandwidth. If routers use different values:

• Router A (ref BW = 100 Mbps): Sees 1 Gbps link as cost 1
• Router B (ref BW = 100 Gbps): Sees 1 Gbps link as cost 100

These inconsistent views cause routing asymmetry and can lead to:

• Suboptimal path selection
• Traffic loops in extreme cases
• Difficult-to-debug routing issues

Best Practices:

1. Document the standard: Include reference bandwidth in network documentation
2. Template configurations: Use configuration templates that include reference bandwidth
3. Audit regularly: Periodic audits ensure all routers maintain consistent settings
4. Start high: Setting reference bandwidth higher than currently needed provides room for future upgrades

Understanding exactly how bandwidth translates into routing decisions requires tracing through the complete calculation process. Let's examine this with a practical network topology.

Scenario Setup:

• Reference Bandwidth: 10,000 Mbps (10 Gbps)
• R1 to R2: 1 Gbps (1000 Mbps)
• R1 to R3: 100 Mbps
• R2 to R4: 10 Gbps (10000 Mbps)
• R3 to R4: 10 Gbps (10000 Mbps)

Step 1: Calculate Individual Interface Costs

R1-R2 Cost = 10000 / 1000 = 10
R1-R3 Cost = 10000 / 100 = 100
R2-R4 Cost = 10000 / 10000 = 1
R3-R4 Cost = 10000 / 10000 = 1

Step 2: Calculate Total Path Costs

Path via R2: R1→R2 + R2→R4 = 10 + 1 = 11
Path via R3: R1→R3 + R3→R4 = 100 + 1 = 101

Step 3: Path Selection

OSPF selects the path with the lowest total cost:

• Path via R2: Cost 11 ✓ (Selected)
• Path via R3: Cost 101 ✗ (Higher cost, not selected)

Despite both paths having 2 hops, OSPF correctly identifies the higher-bandwidth path through R2.

While OSPF uses a simple division formula, EIGRP (Enhanced Interior Gateway Routing Protocol) incorporates bandwidth into a more complex composite metric. Understanding EIGRP's approach reveals different philosophies in metric design.

Key Differences from OSPF:

1. Minimum Bandwidth Focus EIGRP uses the minimum bandwidth along the entire path, not the sum. This reflects the bottleneck principle: a path's effective throughput is limited by its slowest link.

2. Scaled Values EIGRP scales bandwidth using 10^7 / bandwidth_in_kbps, producing larger numbers that provide finer granularity.

3. Composite Nature Bandwidth is combined with delay (and optionally load and reliability) to produce a single metric value.

4. K Values The K values allow administrators to weight different components. By default, only bandwidth (K1) and delay (K3) contribute.

A critical distinction that causes frequent confusion: routing protocols use configured bandwidth, not measured or actual bandwidth. Understanding this difference is essential for correct network design and troubleshooting.

Why This Matters: The Subrate Circuit Scenario

Consider a common scenario: You have a 1 Gbps physical Ethernet port connected to a service provider circuit that only provides 100 Mbps of actual capacity (a common 'subrate' offering).

Consequences of Incorrect Bandwidth Configuration:

1. Overcommitment: If the router believes it has a 1 Gbps path when only 100 Mbps exists, it may prefer this path inappropriately, leading to congestion and packet loss.

2. QoS Miscalculation: Many QoS mechanisms use configured bandwidth to calculate queue sizes and shaping rates. Incorrect bandwidth leads to suboptimal QoS behavior.

3. Routing Suboptimality: Other paths that appear more expensive may actually offer better throughput.

Common Scenarios Requiring Manual Bandwidth Configuration:

• Subrate provider circuits (100 Mbps over GigE interface)
• Frame Relay / ATM virtual circuits
• VPN tunnels with bandwidth guarantees
• DSL/cable with asymmetric speeds
• Wireless backhauls with variable capacity

While automatic cost calculation based on bandwidth provides excellent default behavior, network engineers frequently need to override these values for traffic engineering purposes. Manual cost configuration enables:

• Policy-based routing: Prefer certain paths for business/contractual reasons
• Load distribution: Spread traffic across multiple paths
• Backup designation: Force certain links to be used only during failures
• Metered link avoidance: Reduce usage on expensive pay-per-bit connections

ECMP Load Balancing via Equal Costs:

By configuring equal costs on parallel links, you can enable Equal-Cost Multi-Path (ECMP) routing:

With equal total costs (150 via R2, 150 via R3), OSPF installs both paths in the routing table and load-balances traffic between them. This doubles effective bandwidth while providing redundancy.

Traffic Engineering Scenario:

Consider an organization with two WAN links:

• Primary: 10 Gbps fiber (flat-rate billing)
• Backup: 1 Gbps (metered, $0.02 per GB)

Goal: Use primary exclusively unless failed, then failover to backup.

Implementation:

While bandwidth-based metrics represent a significant improvement over hop count, they carry their own limitations that network engineers must understand.

The Elephant Flow Problem:

Consider this scenario with ECMP (Equal-Cost Multi-Path) routing:

With two equal-cost paths, ECMP splits traffic between them. But if traffic consists of:

• Path A: One large 8 Gbps flow (elephant flow)
• Path B: Many small flows totaling 2 Gbps

Path A becomes congested while Path B is underutilized. Bandwidth metrics provide no mechanism to detect or correct this imbalance—both paths appear equally good.

Modern Approaches:

These limitations have driven development of more sophisticated traffic engineering approaches:

1. Segment Routing: Explicit path control independent of IGP metrics
2. SD-WAN: Real-time path selection based on application requirements and measured link quality
3. Flow Telemetry: Visibility into actual traffic distribution for manual optimization
4. MPLS-TE: Traffic Engineering extensions that consider bandwidth reservations

We've thoroughly explored bandwidth as a fundamental routing metric—from its conceptual foundation through protocol-specific implementations, configuration requirements, and inherent limitations. Let's consolidate the essential knowledge:

Looking Ahead:

Bandwidth represents capacity, but network performance depends on more than raw throughput. In the next page, we'll explore delay as a routing metric—how propagation time, transmission time, and queuing contribute to latency, and how protocols incorporate delay into path selection decisions.

The combination of bandwidth and delay begins to paint a more complete picture of path quality, moving us toward the composite metrics that modern protocols employ for optimal routing.