Link Operational Intelligence (LOI)

AI-Aware Direct-Link Health Exchange for Preemptive Routing Intelligence

Companion Technology Paper

Author: Christopher Soans
Date: August 2026

Download PDF

Figure 1. Link Operational Intelligence (LOI)

Executive Summary

This paper proposes Link Operational Intelligence (LOI), a secure, optional, link-local capability for AI-aware routers and switches to exchange operational health information about their directly connected relationship. LOI complements conventional routing protocols rather than replacing or modifying them. OSPF and IS-IS remain authoritative for topology and reachability, while LOI supplies rapid bidirectional knowledge about the actual forwarding condition of an adjacency.

An AI-aware device can already observe its own queues, drops, utilization, errors, forwarding ASIC/NPU behavior, CPU punts and other telemetry. LOI allows the peer at the opposite end of the link to communicate corresponding observations. Local AI processors can correlate both viewpoints to identify developing physical, congestion, QoS or forwarding-path problems before a routing adjacency necessarily fails.

Backward compatibility is fundamental. LOI is used only when both directly connected devices securely establish LOI capability. A legacy peer simply does not participate; normal Ethernet/IP and routing-protocol operation continues unchanged. No OSPF or IS-IS standards update is required for the core concept.

1. Problem Addressed

A routing adjacency can remain fully established even while the underlying forwarding relationship is deteriorating. Traditional link-up/link-down state and configured routing metrics do not necessarily reveal persistent queue pressure, remote-side packet drops, excessive CPU punts, slow-path forwarding, class-specific QoS congestion, optical degradation or other conditions that can materially impair traffic.

Furthermore, each endpoint normally has only a partial view. A transmitting router may see clean transmission statistics while the receiving router is experiencing queue exhaustion or forwarding-resource pressure. LOI creates a mechanism for the two endpoints to share and correlate these observations.

2. Architectural Role

Function	Primary Question	Mechanism
OSPF/IS-IS	Is the topology reachable and what is its protocol state?	Standards-based link-state routing
LOI	How healthy is this directly connected forwarding relationship from both ends?	Secure link-local operational exchange
AI Intelligence/Control Network	How healthy is the wider network and its candidate paths?	Network-wide telemetry and controller correlation

The three functions remain separable. LOI can provide local value without requiring network-wide AI routing, while a broader AI controller can consume LOI-derived health state as an additional corroborated signal.

3. AI-Aware Router OS Requirements

LOI depends on an AI-aware network operating system that exposes trusted hardware and software telemetry to the local AI processor and controls what information may be shared with the adjacent peer.

• Per-interface transmit and receive utilization, errors, discards and link events.
• Hardware queue occupancy, buffer pressure, drops and persistent microburst indicators.
• QoS class/queue health, including shaping, policing or class-specific delay where measurable.
• ASIC/NPU forwarding state versus CPU punt, exception or process/slow-path behavior.
• Optical or physical-layer health where supported by the interface hardware.
• Local latency, loss or active-probe observations associated with the adjacency.
• Time synchronization, freshness markers and provenance for exchanged observations.
• Policy controls governing which telemetry fields may leave the device.

4. LOI Neighbor Capability Discovery

When an interface becomes operational, an AI-aware device may attempt LOI capability discovery. If the directly connected peer supports LOI, the devices establish an authenticated local relationship and negotiate supported telemetry classes, version, security parameters and update behavior.

If no compatible response is received, the interface is treated as a conventional adjacency. LOI failure must never prevent ordinary Layer 2, Layer 3, OSPF, IS-IS, BGP or other standards-based operation.

Local Device	Peer	Behavior
AI-aware	AI-aware and authenticated	LOI session eligible
AI-aware	Legacy/non-LOI	Conventional operation; no LOI dependency
AI-aware	LOI-capable but authentication fails	Reject LOI; conventional routing remains available subject to normal security policy

5. Secure Link-Local Exchange

LOI should be scoped to the directly connected relationship and should not function as a general routable telemetry protocol by default. Messages should be authenticated, integrity protected, replay resistant and bound to the participating device/interface identities.

A peer should report only authorized operational information relevant to the shared adjacency or its immediate forwarding treatment. This limits the blast radius of compromised telemetry and makes it difficult for a remote system to impersonate a local interface and induce false health decisions.

6. Bidirectional Health Correlation

The principal value of LOI is not merely exchanging counters; it is correlating independent observations. The local AI processor can combine its own measurements with the remote endpoint's measurements and historical baseline.

Local Observation	Remote Observation	Possible Interpretation
TX rate rising; local queues healthy	RX queue and drops rising	Remote receive-side congestion
TX/physical errors rising	RX/CRC errors rising	Probable physical/link degradation
Traffic normal	CPU punts and processing delay rising	Remote forwarding-path impairment
Total utilization moderate	Specific QoS queue saturated	Traffic-class-specific impairment
Short queue spike	Remote state normal shortly afterward	Likely transient microburst; avoid route reaction

7. Link Operational Health State

LOI can maintain a compact operational state derived from correlated telemetry rather than forcing routing systems to react to every counter change. A representative state model is Normal, Elevated, Degraded, Critical and Failed.

Transitions should use persistence thresholds, hysteresis, confidence and minimum hold times. For example, a two-millisecond queue spike should normally be treated differently from sustained queue growth accompanied by rising drops and remote CPU punts.

8. Preemptive Notification

An important objective is to communicate deterioration before the link becomes unusable. A device detecting sustained degradation can send an authenticated LOI health notification to its peer. The receiving AI processor can begin evaluating alternatives without immediately changing routing.

A progression from Elevated to Degraded can trigger local analysis, additional measurements, controller notification or shadow evaluation of alternate paths. Critical state may permit a policy-approved routing or traffic-engineering response after deterministic validation. Physical failure remains handled by ordinary interface, BFD and routing-protocol mechanisms.

9. Interaction with Routing Decisions

LOI must not directly rewrite the OSPF/IS-IS LSDB or originate fabricated protocol information. Instead, LOI-derived state is supplied to the AI-aware OS routing-policy layer. If a response is justified, the system can use an approved deterministic mechanism such as Segment Routing/traffic-engineering policy, constrained steering, ECMP adjustment, controller-installed policy or carefully governed metric action.

This preserves the separation between operational intelligence and protocol truth: LOI identifies a forwarding-quality problem; established routing mechanisms execute any permitted response.

10. AI-Aware Path Constraint

For telemetry to influence an end-to-end routing decision, the broader AI-assisted routing architecture should retain the previously defined AI-aware path constraint. LOI can establish high-confidence health for the immediate adjacency, but it cannot make an opaque downstream legacy segment observable.

Therefore, a local LOI condition may always be used for diagnostics and protection of the directly connected device itself, while network-wide telemetry-derived path preference should be applied only where the required forwarding path is sufficiently AI-aware and observable. Mixed or legacy paths continue to rely on conventional routing information.

11. Relationship to the AI Intelligence/Control Network

LOI provides immediate peer-to-peer knowledge even if the centralized or regional AI controller is temporarily unreachable. When the AI Intelligence/Control network is available, both endpoints can also report their observations and correlated LOI state to the controller.

This enables multi-source corroboration: local counters, remote peer reports, LOI state, controller telemetry and historical baselines can be compared before a significant network-wide action is taken. The dedicated control network also remains useful when production forwarding is congested.

12. Example Scenario: Remote Queue and Punt Degradation

R1 and R2 are directly connected and have an authenticated LOI session. R1 observes normal transmission, no local drops and moderate utilization. R2 observes increasing receive-queue occupancy, rising drops and an abnormal increase in packets punted from hardware forwarding to its CPU.

R2 first advertises an Elevated LOI state. R1 records the condition but makes no routing change. As queue occupancy remains high and CPU punts continue to increase, R2 reports Degraded state with supporting measurements and freshness information. R1's local AI processor correlates this with its own clean transmit side, identifying the impairment as likely remote forwarding degradation rather than a local physical fault.

Both devices report the event through the AI Intelligence/Control network. If an alternate fully AI-aware path is healthy, the controller or local deterministic policy can validate and steer appropriate traffic away from the affected adjacency. OSPF/IS-IS adjacency state need not be falsified or withdrawn merely to represent the performance problem.

13. Avoiding Instability

• Do not treat instantaneous queue depth as a routing metric.
• Use persistence, hysteresis, dampening and change-rate limits.
• Correlate multiple signals before classifying degradation.
• Distinguish physical faults, congestion, QoS impairment and CPU/forwarding impairment.
• Model whether rerouting would overload an alternate path.
• Prefer traffic-class-specific steering when only one QoS class is affected.
• Require deterministic validation and rollback for automated forwarding changes.

14. Failure and Fallback Behavior

• Loss of the LOI session does not bring down the routing adjacency by itself.
• Loss of the AI processor falls back to conventional routing behavior.
• Stale LOI information expires and cannot indefinitely influence decisions.
• Conflicting local and remote telemetry lowers confidence and may trigger additional validation rather than immediate action.
• A legacy replacement device automatically results in non-LOI operation on that adjacency.
• Operators can disable LOI globally or per interface without changing basic routing interoperability.

15. Security Considerations

• Mutual authentication of directly connected LOI peers.
• Message integrity, replay protection and freshness validation.
• Interface/device identity binding.
• Rate limiting to prevent telemetry storms or control-plane exhaustion.
• Authorization of shareable telemetry fields.
• Protection against falsified health reports intended to divert traffic.
• Audit logging of received health states and resulting decisions.
• Independent controller corroboration for high-impact actions where practical.

16. Deployment Model

• Phase 1 — AI-aware OS exposes local interface telemetry and maintains LOI state internally.
• Phase 2 — Enable authenticated LOI discovery and read-only peer telemetry exchange.
• Phase 3 — Operate LOI in advisory mode and compare diagnoses with existing monitoring.
• Phase 4 — Feed LOI state into the AI Intelligence/Control network for correlation.
• Phase 5 — Permit limited, policy-approved deterministic responses on fully AI-aware paths.
• Phase 6 — Evaluate vendor-neutral standardization of LOI discovery and telemetry semantics if operational benefits are demonstrated.

17. Benefits

• Provides visibility into both ends of a directly connected forwarding relationship.
• Detects degradation that may not cause a routing adjacency failure.
• Allows preemptive warning before queue, forwarding or physical conditions become critical.
• Improves diagnosis by correlating local and remote observations.
• Separates high-frequency operational intelligence from OSPF/IS-IS flooding.
• Preserves backward compatibility with legacy routers.
• Provides a local intelligence layer that remains useful independently of centralized AI control.
• Strengthens network-wide AI decisions through corroborated link-health data.

18. Scope Boundary

LOI is not a replacement for OSPF, IS-IS, BFD, Ethernet OAM or existing physical-layer diagnostics. It is an AI-aware correlation and peer-intelligence layer that can consume such information and securely communicate selected operational state across a directly connected adjacency.

The proposal does not require an OSPF or IS-IS standards update, does not permit AI-generated information to enter the authoritative LSDB as if it were protocol truth, and does not assume that legacy devices expose telemetry they cannot provide.

19. Relationship to the Companion Architecture

Technology Concept	Role
AI-Accelerated Link-State Routing Infrastructure	Faster local processing of verified OSPF/IS-IS LSDB information
AI-Assisted Routing Intelligence	Network-wide telemetry-informed path intelligence over the AI Intelligence/Control network
Link Operational Intelligence (LOI)	Secure, bidirectional operational intelligence between directly connected AI-aware devices

Together, the three concepts create distinct but complementary layers: deterministic topology processing, immediate adjacent-link intelligence and network-wide forwarding intelligence.

20. Conclusion

Link Operational Intelligence provides a practical way for AI-aware network devices to understand the real operational condition of their directly connected links without redefining OSPF or IS-IS. By securely exchanging peer telemetry and correlating both sides of an adjacency, routers can recognize developing congestion, QoS impairment, forwarding slow-path behavior and physical degradation before those conditions necessarily become protocol failures.

The architecture remains conservative where it matters: routing protocols retain authority, legacy peers continue to function normally, LOI information expires when stale, and any routing response passes through deterministic policy and validation. This makes LOI a natural companion to AI-accelerated LSDB processing and network-wide AI-assisted routing intelligence.

© 2026 Christopher Soans. All rights reserved.

This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

Related Posts

Loading related posts...