Interface Health

Interface and link health assessment with error counter analysis, optical power monitoring, discard diagnosis, and utilization trending. Multi-vendor coverage for Cisco IOS-XE/NX-OS, Juniper JunOS, and Arista EOS with severity-tiered thresholds for physical and data-link layer metrics.

Interface Health Assessment

Threshold-driven diagnostic skill for interface and link health. Covers the physical and data-link layers — error counters, optical power levels, discard rates, interface flaps, and bandwidth utilization. Each metric is evaluated against four severity tiers (Normal / Warning / Critical / Emergency) with vendor-specific collection commands.

Commands are labeled [Cisco], [JunOS], or [EOS] where syntax diverges. Unlabeled statements apply to all three vendors. Detailed command syntax is in references/cli-reference.md; full threshold tables with per-optic-type ranges are in references/threshold-tables.md.

When to Use

• Interface reported down or flapping (repeated up/down transitions)
• Users reporting packet loss or degraded throughput on a link
• Monitoring alerts for CRC errors, input errors, or output drops
• Pre/post maintenance validation of link quality after cable or optic swap
• Optical power alarms from DOM (Digital Optical Monitoring) readings
• Capacity planning — identifying interfaces approaching saturation
• Troubleshooting latency spikes that correlate with interface congestion
• Baseline collection for new link turn-ups or circuit migrations

Prerequisites

• SSH or console access to the device (read-only privilege sufficient)
• Interfaces to evaluate are identified (specific interfaces or all active)
• Baseline error counts or a prior snapshot for delta comparison — without a baseline, only instantaneous rates and absolute counters are available
• Knowledge of expected link parameters: speed, duplex, media type (copper vs fiber), SFP model, and cable distance
• For optical checks: SFP/QSFP modules with DOM support installed

Procedure

Work through each step sequentially. Early steps collect broad status; later steps drill into specific failure domains identified by prior output.

Step 1: Interface Status Overview

Collect admin state, operational state, speed, duplex, and media type for all interfaces under review.

[Cisco]

text

show interfaces status
show interfaces [intf] | include line protocol|BW|duplex

[JunOS]

text

show interfaces terse
show interfaces [intf] | match "Physical|Speed|Duplex|Link-level"

[EOS]

text

show interfaces status
show interfaces [intf] | include line protocol|BW|duplex

Record each interface: name, admin/oper state, speed, duplex, media type. Any interface that is admin up but operationally down requires immediate investigation — skip to the Decision Trees section for that interface. Duplex mismatches (one end full, other half) cause late collisions and must be resolved before error analysis is meaningful.

Step 2: Error Counter Analysis

Collect error counters and calculate per-interval rates. Raw counters are cumulative since last clear — always compute a delta over a known interval (minimum 5 minutes) for actionable rates.

[Cisco]

text

show interfaces [intf] | include CRC|input errors|output errors|frame|runts|giants
show interfaces [intf] counters errors

[JunOS]

text

show interfaces [intf] extensive | match "CRC|Errors|Framing|Runts|Giants"

[EOS]

text

show interfaces [intf] counters errors
show interfaces [intf] | include CRC|input errors|output errors|runts|giants

Key counters to evaluate:

• CRC errors — corrupted frames; indicates physical-layer problems (bad cable, dirty fiber, failing SFP, EMI)
• Input errors — superset including CRC, frame, overrun; aggregate indicator of receive-path health
• Output errors — transmission failures; often buffer exhaustion or interface congestion
• Frame errors — non-integer-octet frames; typically duplex mismatch or bad NIC
• Runts — undersized frames (<64 bytes); usually collision fragments or bad NIC
• Giants — oversized frames; MTU mismatch between endpoints

Compare rates against thresholds in references/threshold-tables.md. Any counter incrementing steadily (not a stale historical value) at Warning level or above warrants investigation.

Step 3: Discard Analysis

Evaluate input and output discards separately — they have different root causes.

[Cisco]

text

show interfaces [intf] | include drops|discard|queue
show policy-map interface [intf]

[JunOS]

text

show interfaces queue [intf]
show class-of-service interface [intf]

[EOS]

text

show interfaces [intf] counters discards
show qos interface [intf]

• Output discards — interface transmit ring full. Causes: sustained congestion (traffic exceeds link capacity), inadequate QoS scheduling, microbursts overwhelming shallow buffers.
• Input discards — receive ring full. Causes: CPU unable to process at line rate (control plane punt), input QoS policer drops, or receive buffer exhaustion.
• Queue drops — per-queue drops visible in QoS policy output. Identify which traffic class is affected to prioritize remediation.

High output discards with low utilization suggests microburst activity — short-duration traffic spikes that don't appear in 5-minute utilization averages but overflow interface buffers.

Step 4: Interface Reset and Flap Detection

Identify interfaces with recent or recurring resets.

[Cisco]

text

show interfaces [intf] | include resets|Last input|Last output|last change

[JunOS]

text

show interfaces [intf] extensive | match "Last flapped|Resets"

[EOS]

text

show interfaces [intf] | include resets|Last input|Last output
show logging | include [intf].*up|[intf].*down

Record reset counts and last-flap timestamps. Correlate flap events with error counter spikes — a link that flaps and accumulates CRC errors on recovery likely has a physical-layer issue (loose cable, marginal SFP). Frequent resets without errors may indicate auto-negotiation failures or spanning-tree reconvergence triggers.

Threshold: >3 resets/hour is Critical; >10 resets/hour is Emergency. See references/threshold-tables.md for full severity tiers.

Step 5: Optical Power Monitoring

For fiber interfaces with DOM-capable SFPs, collect Tx power, Rx power, laser bias current, and module temperature.

[Cisco]

text

show interfaces [intf] transceiver detail

[JunOS]

text

show interfaces diagnostics optics [intf]

[EOS]

text

show interfaces [intf] transceiver detail

Key readings:

• Tx Power (dBm) — transmit optical power. Out-of-range indicates SFP degradation or failure.
• Rx Power (dBm) — received optical power. Low Rx with normal Tx on the remote end indicates fiber attenuation (dirty connector, bend loss, distance exceeded, bad splice).
• Laser Bias Current (mA) — current driving the laser. Rising bias over time indicates SFP aging; high bias with low Tx power means the SFP is compensating for degradation.
• Temperature (°C) — module operating temperature. Elevated temperature accelerates SFP aging and can cause transmission errors.

Compare readings against the per-optic-type tables in references/threshold-tables.md. The tables provide manufacturer spec ranges for common SFP types (1G-SX, 10G-SR, 10G-LR, 25G-SR, 100G-SR4).

Step 6: Utilization Assessment

Measure bandwidth usage to identify congested or underutilized links.

[Cisco]

text

show interfaces [intf] | include input rate|output rate|reliability
show interfaces [intf] summary

[JunOS]

text

show interfaces [intf] traffic
show interfaces [intf] statistics traffic

[EOS]

text

show interfaces [intf] | include input rate|output rate
show interfaces [intf] counters rates

Calculate utilization as a percentage of interface speed. Note that CLI "input/output rate" values are typically 5-minute weighted averages — they smooth out microbursts. For burst detection, correlate with output discards (Step 3) and use streaming telemetry or shorter polling intervals if available.

Threshold Tables

Summary of key thresholds used in this skill. Full per-optic-type tables and detailed severity definitions are in references/threshold-tables.md.

Metric	Normal	Warning	Critical	Emergency
CRC errors/5min	0	1–5	6–50	>50
Input errors/5min	0–2	3–20	21–100	>100
Output discards/5min	0–10	11–100	101–1000	>1000
Interface resets/hr	0	1–2	3–10	>10
Rx Power vs low-warn	>3 dBm margin	1–3 dBm margin	0–1 dBm margin	Below low-alarm
Utilization %	0–50%	51–75%	76–90%	>90%

Decision Trees

High Error Rate

text

Errors incrementing (CRC, input, frame)
├── Single interface affected?
│   ├── Yes → Physical layer issue
│   │   ├── Fiber interface?
│   │   │   ├── Check Rx power → Low? → Clean connectors, check fiber run
│   │   │   ├── Rx power normal? → Check SFP seating, try reseat
│   │   │   └── SFP temperature high? → Check airflow, replace SFP
│   │   └── Copper interface?
│   │       ├── Check cable length (<100m for Cat6)
│   │       ├── Check for EMI sources near cable run
│   │       └── Swap cable, test with known-good
│   └── No → Multiple interfaces affected
│       ├── Same line card/module? → Suspect line card hardware
│       ├── Same patch panel? → Suspect patch panel or cable tray
│       └── All interfaces? → Upstream power or grounding issue
├── Frame errors specifically?
│   └── Check duplex settings both ends → Mismatch causes late collisions
├── Giants specifically?
│   └── Check MTU both ends → Must match; check for jumbo frame config
└── Runts specifically?
    └── Check for collision domain issues or bad NIC on connected device

High Discards

text

Discards incrementing
├── Output discards?
│   ├── Utilization >75%? → Congestion — consider link upgrade or ECMP
│   ├── Utilization <50% but discards high? → Microburst activity
│   │   ├── Check per-queue drops → Identify affected traffic class
│   │   └── Increase interface buffer or tune QoS scheduling
│   └── QoS policy applied?
│       ├── Check policer/shaper rates → May be too restrictive
│       └── Check queue allocation → Priority queue starving best-effort
├── Input discards?
│   ├── CPU-bound traffic? → Check CoPP counters for punt drops
│   ├── Input policer configured? → Check rate vs actual traffic
│   └── Receive buffer exhaustion → Rare; check for flood traffic
└── Intermittent vs sustained?
    ├── Sustained → Capacity issue; plan upgrade
    └── Intermittent → Burst-driven; tune buffers or QoS

Optical Power Out of Range

text

Optical reading outside normal range
├── Rx power low?
│   ├── Remote Tx power normal? → Fiber path issue
│   │   ├── Clean both connectors (IPA wipes, not compressed air alone)
│   │   ├── Check for fiber bends exceeding minimum bend radius
│   │   ├── Verify distance within SFP spec (e.g., 10G-SR max 300m OM3)
│   │   └── Test with OTDR if available → Locate fault point
│   └── Remote Tx power also low? → Remote SFP failing
├── Tx power low?
│   ├── Laser bias current high? → SFP degrading, plan replacement
│   ├── Laser bias current normal? → SFP may be misconfigured
│   └── Temperature elevated? → Fix cooling, then recheck Tx power
├── Rx power high?
│   └── Short fiber run without attenuator → Add inline attenuator
└── Temperature alarm?
    ├── Ambient temp normal? → SFP internal issue, replace
    └── Ambient temp high? → Fix chassis/room cooling first

Report Template

text

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

INTERFACE SUMMARY:
- Total interfaces reviewed: [n]
- Admin Up / Oper Up: [n]
- Admin Up / Oper Down: [n] — [list]
- Interfaces with active errors: [n]
- Interfaces with discards: [n]

FINDINGS:
1. [Severity] [Category] — [Description]
   Interface: [name]
   Observed: [metric value and rate]
   Threshold: [Normal/Warning/Critical/Emergency boundary crossed]
   Root Cause: [diagnosis from decision tree]
   Action: [recommended remediation]

OPTICAL STATUS:
- Interfaces with DOM data: [n]
- Rx power warnings: [list with dBm values]
- Tx power warnings: [list with dBm values]
- SFPs approaching end-of-life (rising bias current): [list]

UTILIZATION SUMMARY:
- Interfaces >75% utilized: [list with peak and average rates]
- Interfaces >90% utilized: [list — immediate attention required]

RECOMMENDATIONS:
- [Prioritized action list]

NEXT CHECK: [EMERGENCY: 1hr, CRITICAL: 4hr, WARNING: 24hr, NORMAL: 7d]

Troubleshooting

CRC Errors on Fiber with Normal Optical Power

Optical power within spec but CRC errors incrementing. Common causes: wavelength mismatch between SFP types (e.g., SX connected to LR), dirty connector on the inside of the SFP cage (not the fiber tip), or SFP incompatibility with the switch (non-qualified optic). Try: clean the SFP receptacle, verify both ends use the same SFP type, test with a vendor-qualified optic.

Output Discards with Low Utilization

Interface shows <30% average utilization but output discards are climbing. This is almost always microburst traffic — sub-second spikes that exceed link capacity during the burst but average out below the utilization threshold. Diagnose with: per-queue drop counters (shows which traffic class), interface buffer allocation stats. Remediate with: QoS scheduling adjustments, increased interface buffer depth, or traffic shaping at the ingress point.

Interface Stuck in Down/Down After Cable Swap

Admin up, operationally down after replacing a cable or SFP. Check: SFP is fully seated (push firmly until click), fiber polarity is correct (Tx-to-Rx crossover), SFP type matches remote end, speed/duplex is set to auto or matches. On [Cisco], check show interfaces [intf] | include err-disabled — the port may have been error-disabled by a protection feature (BPDU guard, UDLD, link-flap detection). Recover with shutdown / no shutdown after fixing the root cause.

Flapping Interface with No Errors

Interface cycles up/down every few seconds with zero error counters. This suggests a negotiation or protocol issue, not a physical fault. Common causes: auto-negotiation incompatibility (force speed/duplex on both ends), STP topology changes causing repeated blocking/forwarding transitions, UDLD aggressive mode detecting unidirectional link. Check spanning-tree state and UDLD status on the interface.

Rising Laser Bias with Stable Tx Power

Laser bias current increasing over weeks/months while Tx power remains stable. The SFP is compensating for laser degradation by driving more current. This is normal aging but indicates the SFP will eventually fail — Tx power will drop when the laser can no longer compensate. Plan proactive replacement before the Tx power begins declining. Track the trend: if bias current exceeds 80% of the manufacturer's max specification, schedule replacement within 30 days.