RETICUX BGP Mastery — Day 023 — BGP Multipath: Turning Multiple Equal Paths into Forwarding Capacity

Learning objective

Distinguish best-path selection from multipath installation and configure controlled iBGP/eBGP multipath on IOS XE.

BGP Multipath: Turning Multiple Equal Paths into Forwarding Capacity

---

1. Opening — the decision point that is easy to misread

BGP best-path selection is sequential. A router does not assign a single universal “route score” and then pick the smallest or largest number. It evaluates eligible paths through an ordered decision process. The first meaningful difference can end the comparison; later attributes may never be examined.

That matters operationally because engineers often see two routes and jump directly to AS_PATH length. By the time AS_PATH is reached, several earlier decisions may already have eliminated one candidate. Conversely, when two routes remain equal through MED, later implementation-specific or topology-dependent decisions become decisive.

This day isolates one such decision point so that the result can be proven rather than inferred from a route table alone.

2. Standards behavior versus Cisco behavior

RFC 4271 defines BGP's route selection framework but deliberately leaves implementation-specific selection details to implementations. Cisco IOS XE documents an ordered decision process containing Cisco-local and BGP attributes.

The engineering rule is therefore: use RFC text to understand protocol semantics, and Cisco documentation to verify the actual IOS XE decision order and configuration knobs. Do not copy an algorithm from a different vendor and assume Cisco behaves identically.

This distinction becomes particularly important for router ID, multipath, best-external and route-reflector features.

3. Scenario

R1 receives two or more paths that qualify for multipath after the relevant BGP selection conditions are satisfied. The lab first demonstrates the default single-best-path behavior, then enables a controlled number of parallel paths and verifies that multiple paths are installed.

The lab uses documentation-safe addressing and private lab ASNs. No production prefixes, credentials or real operator identifiers are used.

4. Topology

                         AS 65100
                 +---------------------+
                 |       R1 / Core     |
                 |   BGP decision point|
                 +----------+----------+
                            | iBGP
                            |
                         +--+--+
                         | R2  |
                         +--+--+
                            |
                 +----------+----------+
                 |                     |
              eBGP                  eBGP
                 |                     |
              +--+--+               +--+--+
              | ISP-A|               | ISP-B|
              |65110 |               |65120 |
              +-----+               +-----+

Prefix under test: 203.0.113.0/24

5. Prerequisites

• IOS XE 17.18.x target image or equivalent supported IOS XE 17.x lab image.
• Reachable loopbacks/interfaces before BGP policy is tested.
• IPv4 unicast address family enabled.
• Private/documentation-safe ASNs and prefixes.
• show ip bgp, show ip bgp summary, and show ip route available.
• NTP or a stable lab clock is recommended for incident timestamps.

6. Baseline configuration

router bgp 65100
 address-family ipv4 unicast
  maximum-paths ibgp 2
 exit-address-family

The configuration above is intentionally scoped to the learning objective. It should not be described as a universal production template.

7. Verification before modification

Check show ip bgp 203.0.113.0 for multiple eligible paths and show ip route 203.0.113.0 for multiple installed next hops. Confirm that the paths satisfy the platform's multipath criteria rather than assuming equal prefix reachability is sufficient.

Record the baseline best path before changing the single variable under test. The evidence must show both the BGP table and the installed IP route when forwarding behavior is part of the objective.

8. Controlled modification

Enable only the appropriate multipath command for the topology. First prove that BGP has one installed best path. Then enable two-path multipath and compare the BGP table with the IP routing table.

Change only the variable under investigation. Do not simultaneously alter LOCAL_PREF, MED, AS_PATH, next-hop reachability and multipath settings; doing so destroys causal clarity.

9. Fault injection

Illustrative lab — not a real incident.

Illustrative lab — not a real incident. Configure two paths that look equal to the operator but differ in an attribute required for multipath. The symptom is that only one path is installed even though both paths appear in the BGP table.

The purpose of the fault is to create a recognizable symptom while preserving enough evidence to identify the exact decision point.

10. Troubleshooting

Use this evidence chain:

1. Confirm the affected prefix.
2. Confirm both candidate paths are present.
3. Compare attributes in decision order.
4. Confirm the next hop is recursively reachable.
5. Identify the first attribute where the candidates differ.
6. Confirm whether the result is a best-path decision or a multipath/install decision.
7. Verify the selected route in the RIB.
8. Perform a positive forwarding test.
9. Perform a negative/containment test where safe.
10. Record the smallest proven cause.

Useful IOS XE commands

show ip bgp 203.0.113.0
show ip bgp 203.0.113.0 longer-prefixes
show ip bgp summary
show ip route 203.0.113.0
show ip route <next-hop>
show ip bgp neighbors <peer> advertised-routes
show ip bgp neighbors <peer> routes

Adapt the command set to the actual feature under test. Do not claim output was observed unless the exact lab was executed.

11. Root cause

The cause is the difference between path eligibility, best-path selection and multipath installation. Multiple BGP paths can exist without all of them being installed in the forwarding table.

12. Post-fix verification

Re-run the same evidence set used before the change. The comparison should demonstrate the intended decision change without unrelated routing changes.

13. Rollback

Remove maximum-paths or restore the previous value. Confirm that only the best path remains installed and that forwarding is stable.

14. Production lessons

Multipath is a capacity and resiliency feature, not simply a second best-path algorithm. Treat memory, hardware path limits, hashing behavior and convergence as part of the design.

15. Knowledge check

1. Question: What is the first decision point that can distinguish the two candidate paths in this lab?

• Answer: Inspect the ordered attributes and identify the first actual difference; do not assume AS_PATH is always the first useful discriminator.

1. Question: Why must a best-path change be verified in both the BGP table and the IP routing table?

• Answer: A BGP path can be selected while recursive next-hop or installation conditions prevent the expected forwarding entry.

1. Question: What configuration change would make the lab result misleading?

• Answer: Changing multiple selection inputs simultaneously, because the engineer can no longer prove which factor caused the outcome.

Engineering notes

Cisco documentation notes additional memory use for iBGP multipath and platform-dependent limits. Do not publish a universal maximum without checking the exact platform and release.

16. Sources

• RFC 4271 — IETF/RFC Editor; standards baseline for BGP behavior relevant to this post.
• Cisco IOS XE 17.x iBGP Multipath Load Sharing — Cisco official configuration/implementation documentation for IOS XE 17.x/17.18.x.
• RFC 4456 where route-reflector behavior or Cluster List is discussed.
• RFC 7911 where ADD-PATH behavior is discussed.

Access date: 12 August 2026