BGP Protocol Analysis#

Protocol-reasoning-driven analysis skill for BGP peering, path selection, and route propagation. Unlike device health checks that compare metrics against thresholds, BGP analysis requires interpreting protocol state machines, walking the best-path algorithm, and validating policy application across the control plane.

Commands are labeled [Cisco], [JunOS], or [EOS] where syntax diverges. Unlabeled statements apply to all three vendors.

When to Use#

• BGP peer reported down or stuck in a non-Established state
• Suspected route leak — prefixes appearing in tables where they should not
• Path selection not matching expectations after policy changes
• Convergence too slow after planned maintenance or unplanned failover
• Post-change verification of BGP configuration (new peers, policy updates, community changes)
• Capacity planning for prefix table growth or session scaling
• Investigating asymmetric routing caused by inconsistent BGP attributes

Prerequisites#

• SSH or console access to the router (read-only privilege sufficient)
• BGP process running on the device with at least one configured peer
• Knowledge of expected peer topology: which neighbors should be up, expected prefix counts per peer, intended path selection outcomes
• Awareness of configured routing policy: route-maps, prefix-lists, community filters, AS-path access lists, and local-preference assignments
• For iBGP: understanding of the route-reflector or full-mesh topology

Procedure#

Follow this diagnostic flow sequentially. Each step builds on the data from prior steps. The procedure moves from broad inventory to targeted diagnosis.

Step 1: BGP Session Inventory#

Collect all peer states and compare against expected topology.

[Cisco]

text

show bgp ipv4 unicast summary

[JunOS]

text

show bgp summary

[EOS]

text

show ip bgp summary

Record each neighbor: address, AS number, state, prefixes received, up/down time. Compare against expected topology — every configured peer should appear. A peer missing from output means it was never configured or was removed. Any peer not showing a numeric prefix count is not Established — proceed to Step 2 for that peer.

Step 2: Peer State Diagnosis#

For any peer not in Established state, the BGP FSM state reveals the failure domain. This is the core diagnostic reasoning step.

[Cisco]

text

show bgp ipv4 unicast neighbors [addr] | include state|last reset|error

[JunOS]

text

show bgp neighbor [addr] | match "State|Last Error|Last State"

[EOS]

text

show ip bgp neighbors [addr] | include state|last reset|error

Interpret the FSM state to isolate the failure:

• Idle → BGP process not attempting connection. Causes: administratively shut down, no route to peer address, configured remote AS does not match, or maximum-prefix limit hit triggering teardown.
• Connect → TCP SYN sent, waiting for response. Peer is unreachable at Layer 3 or a firewall is blocking TCP port 179.
• Active → TCP connection attempt failed, retrying. Same causes as Connect but the router has cycled back. Check: ACLs blocking port 179, peer not configured for this neighbor, peer address unreachable.
• OpenSent → TCP connected, OPEN message sent, no reply. Remote end accepted TCP but is not sending OPEN — typically remote BGP not configured for this neighbor or remote peer in admin shutdown.
• OpenConfirm → OPEN received but parameters rejected. Check: AS number mismatch, capability negotiation failure (AFI/SAFI mismatch), hold timer negotiation failure, authentication (MD5/TCP-AO) mismatch.

Check "last reset reason" and "last error" fields — they often provide the definitive cause. Reference: references/state-machine.md for full FSM detail.

Step 3: Route Table Analysis#

For Established peers, verify prefix exchange matches expectations.

[Cisco]

text

show bgp ipv4 unicast neighbors [addr] routes | include Total

[JunOS]

text

show route receive-protocol bgp [addr] table summary

[EOS]

text

show ip bgp neighbors [addr] received-routes | include Total

Compare received prefix count against baseline. Significant deviation indicates:

• Drop >10% → upstream is withdrawing routes (maintenance, filter change, or failure)
• Increase >10% → route leak or new prefixes originated upstream
• Zero received → peer is Established but sending no routes (missing export policy on JunOS, or outbound filter on remote blocking everything)

Check advertised prefix count similarly — confirm this router is sending the expected number of routes to each peer.

Step 4: Path Selection Verification#

When traffic takes an unexpected path, walk the BGP best-path algorithm to identify which attribute is making the selection.

[Cisco]

text

show bgp ipv4 unicast [prefix] bestpath

[JunOS]

text

show route [prefix] detail | match "AS path|Local|MED|Weight|preference"

[EOS]

text

show ip bgp [prefix] detail

The best-path algorithm evaluates in this order (first difference wins):

1. Weight (Cisco/EOS local, highest wins — JunOS does not use weight)
2. Local Preference (highest wins, default 100)
3. Locally originated (network/aggregate preferred over learned)
4. AS Path length (shortest wins)
5. Origin (IGP < EGP < Incomplete)
6. MED (lowest wins, compared only within same neighbor AS by default)
7. eBGP over iBGP (external preferred)
8. IGP metric to next-hop (lowest wins)
9. Router ID (lowest wins, tiebreaker)

Identify which attribute selects the current best path. If unexpected, check the route-map or policy applying that attribute on ingress.

Step 5: Route Filtering Validation#

Verify that route-maps, prefix-lists, and community filters apply as intended.

[Cisco]

text

show bgp ipv4 unicast neighbors [addr] policy

[JunOS]

text

show policy [policy-name] | display detail

[EOS]

text

show route-map [name]

For suspected route leaks: examine the RIB for prefixes that should not be present. Check inbound filters on the peer that is the source. Common leak causes: missing or misordered prefix-list entry, regex error in AS-path filter, community match that is too broad.

For missing routes: verify the outbound policy on the advertising peer is not filtering the prefix. On JunOS, a peer with no export policy sends nothing by default — this is the most common JunOS-specific omission.

Step 6: Convergence Assessment#

Evaluate convergence behavior and route stability.

[Cisco]

text

show bgp ipv4 unicast dampening dampened-paths

[JunOS]

text

show route damping suppressed

[EOS]

text

show ip bgp dampening dampened-paths

Check for dampened (suppressed) routes — these indicate persistent flapping. Review the BGP update activity: high update/withdrawal rates indicate churn. After a planned change, measure convergence time from the change event to the last BGP update. Compare against the target convergence window.

Threshold Tables#

Operational parameter norms for BGP — these are protocol-level expectations, not device resource thresholds.

ParameterCisco DefaultJunOS DefaultEOS DefaultNotesHold Timer180s90s180sNegotiated to lower valueKeepalive Interval60s30s60sHold/3 by conventionConnectRetry Timer120sVaries120sTime between TCP attemptsMRAI (eBGP)30s0s (immediate)30sMinimum Route Advertisement IntervalMRAI (iBGP)5s0s5sLower than eBGP for faster iBGP convergenceDefault Local Pref100100100Same across vendors

Table Size Norms (IPv4 unicast):

Deployment TypeExpected PrefixesWarningCriticalInternet edge (full table)~950K>1M>1.1MInternet edge (partial)5K–100KVariesPer designEnterprise WAN100–10K>2x baseline>5x baselineData center leaf50–5K>2x baseline>5x baseline

Convergence Targets:

ScenarioTargetAcceptableDegradedeBGP failover< 90s90–180s> 180siBGP reconvergence< 30s30–60s> 60sFull table reload< 5min5–10min> 10min

Decision Trees#

Peer Not Established#

text

Peer not in Established state
├── State: Idle
│   ├── Admin shut? → Check config for "neighbor shutdown" / "deactivate"
│   ├── No route to peer? → Check IGP/static route to peer address
│   ├── Prefix-limit exceeded? → Check logs for max-prefix teardown
│   └── AS mismatch? → Verify "remote-as" matches peer's local AS
│
├── State: Connect / Active
│   ├── Peer reachable? → Ping/traceroute peer address
│   │   ├── No → Fix Layer 3 reachability (IGP, static route)
│   │   └── Yes → TCP port 179 blocked?
│   │       ├── ACL on local device? → Check interface/control-plane ACL
│   │       ├── ACL on remote device? → Check remote inbound ACL
│   │       └── Firewall between peers? → Verify TCP/179 permitted both directions
│   └── Peer configured? → Verify remote has this router as neighbor
│
├── State: OpenSent
│   ├── Remote not sending OPEN → Peer may not be configured for this neighbor
│   ├── TCP resets after connect → Check for TTL issues (eBGP multihop)
│   └── Authentication? → Verify MD5/TCP-AO passwords match both sides
│
├── State: OpenConfirm
│   ├── Capability mismatch? → Check AFI/SAFI (IPv4/IPv6/VPNv4) match
│   ├── AS mismatch in OPEN? → Verify configured AS matches OPEN AS
│   ├── Hold timer = 0 on one side? → Both peers must agree or both use 0
│   └── Check NOTIFICATION message → Decode error code/subcode
│
└── Established but dropping
    ├── Hold timer expiry? → Keepalives not arriving (CPU, QoS, path issue)
    ├── NOTIFICATION received? → Decode error code for root cause
    ├── Route refresh storm? → Peer sending excessive route-refresh requests
    └── Max-prefix limit? → Peer sending more prefixes than limit allows

Unexpected Route Selection#

text

Wrong path selected for prefix
├── Check Weight (Cisco/EOS only)
│   └── Weight set via route-map? → Highest weight wins
├── Check Local Preference
│   └── Local pref differs? → Set via inbound route-map; highest wins
├── Check AS Path Length
│   ├── AS prepending applied? → Verify prepend count
│   └── AS path differs? → Shortest wins
├── Check Origin
│   └── IGP vs Incomplete? → IGP (network statement) preferred
├── Check MED
│   ├── MED comparison enabled across AS? → "always-compare-med"
│   └── MED set correctly? → Lowest wins within same neighbor AS
├── Check eBGP vs iBGP
│   └── External path preferred over internal if equal above
├── Check IGP metric to next-hop
│   └── Closest exit wins (hot-potato routing)
└── All equal → Lowest Router ID wins (or oldest route if stable)

Report Template#

text

BGP ANALYSIS REPORT
====================
Device: [hostname]
Vendor: [Cisco | JunOS | EOS]
Check Time: [timestamp]
Performed By: [operator/agent]

SESSION STATUS:
- Total configured peers: [n]
- Established: [n] | Not Established: [n]
- Peers requiring attention: [list with FSM states]

FINDINGS:
1. [Severity] [Category] — [Description]
   Peer: [neighbor address / AS]
   Observed: [state or metric]
   Expected: [normal state or value]
   Root Cause: [diagnosis from decision tree]
   Action: [recommended remediation]

PATH ANALYSIS:
- Prefix: [prefix under review]
- Selected path via: [next-hop / AS path]
- Selecting attribute: [which best-path attribute decided]
- Expected path: [if different, what was expected and why]

ROUTE TABLE SUMMARY:
- IPv4 prefixes received: [total across all peers]
- Baseline deviation: [% change from expected]

RECOMMENDATIONS:
- [Prioritized action list]

NEXT CHECK: [based on severity — CRITICAL: 4hr, WARNING: 24hr, HEALTHY: 7d]

Troubleshooting#

Session Flapping#

Peer cycles between Established and Idle/Active repeatedly. Common causes: unstable underlying transport (IGP flap, link errors), aggressive hold timers on congested control planes, or MTU issues on the path causing fragmented keepalives to be dropped. Check last reset reason and correlate with interface or IGP events at the same timestamps.

Route Oscillation#

The same prefix alternates between two or more paths. Caused by inconsistent MED comparison across route reflectors, or deterministic-MED not enabled when multiple exit points exist to the same neighbor AS. Enable always-compare-med and deterministic-med to stabilize.

Memory Pressure from Full Table#

Full Internet table (~950K IPv4 prefixes) requires 1–2 GB of RIB memory depending on path diversity. Symptoms: slow convergence, peer resets during table reload. Mitigate with soft-reconfiguration inbound (trades memory for stability) or ORF (Outbound Route Filtering) to reduce inbound load.

Community Stripping#

Routes arrive without expected communities. Check each transit AS in the path — many providers strip non-standard communities by default. Use large communities (RFC 8092) for end-to-end propagation across providers that strip standard communities.

JunOS Default Export Policy#

JunOS sends no routes to a peer without an explicit export policy. If a peer shows Established with zero prefixes sent, add an export policy. This is the most common JunOS-specific BGP issue and does not occur on Cisco or EOS.