A dynamic routing system that continuously monitors real-time network telemetry — latency, bandwidth utilization, CPU load, and throughput — across available paths, and uses rule-based configuration to select and switch to the optimal route as network conditions change. Routes are selected proactively based on measured performance, not reactively after degradation is already affecting traffic.
Traditional routing protocols (OSPF, BGP, EIGRP) select paths based on static cost metrics — hop count, configured bandwidth, administrative distance — that don't reflect real-time network conditions. A route with the lowest hop count may currently have 200ms latency because an intermediate node is under CPU load. The protocol won't switch to an alternative path until the preferred path fails completely, not when it degrades. Traffic experiences the degradation for the entire duration between failure and failover.
The system deploys telemetry agents at each network node to continuously measure and report real-time performance metrics: latency (round-trip time on each available path), bandwidth utilization, CPU load at intermediate nodes, and end-to-end throughput. A central route optimization engine scores each available path across all collected metrics using a weighted formula, then applies rule-based configuration to select the optimal path for each traffic class. Routes switch proactively when the current path's score falls below threshold — before degradation affects traffic quality.
The rule-based configuration layer is what enables the system to optimize differently for different traffic types: latency-sensitive financial transactions are routed over paths scored primarily on latency; bulk data transfers are routed over paths scored primarily on throughput and bandwidth. A single physical network can serve multiple traffic classes simultaneously, each optimized for its specific performance requirements.
Telemetry agents deployed at each network node measure performance on every available path at a configurable sampling interval — typically sub-second for production financial networks. Measurements are timestamped and transmitted to the route optimization engine in real time. The engine maintains a rolling window of recent measurements for each path, computing moving averages that smooth measurement noise while remaining responsive to genuine performance changes.
The four-metric framework captures the full performance picture for a network path. Latency measures user-perceived delay. Bandwidth utilization indicates remaining headroom. CPU load at intermediate nodes predicts impending latency increases before they appear in latency measurements — a node at 95% CPU will soon add queuing delay even if current latency is still acceptable. Throughput measures actual delivered data rate, which may differ from theoretical bandwidth when packet loss or retransmission is occurring.
The route optimization engine computes a composite performance score for each available path using a weighted formula. Weights are specified in the rule-based configuration per traffic class: a latency-sensitive class might weight latency at 60% and bandwidth at 20%; a throughput-sensitive class might weight throughput at 50% and bandwidth at 30%. The engine maintains current scores for all paths and re-evaluates route assignments when scores change beyond a configured hysteresis threshold.
The hysteresis threshold prevents route flapping — the engine won't switch to an alternative path that scores only marginally higher than the current path, because the disruption cost of switching outweighs the marginal performance gain. The threshold is configured per traffic class, allowing latency-sensitive traffic to switch more readily (where even small latency improvements justify a switch) while bulk transfers switch only when the alternative path offers a substantial improvement.
Score = w₁·latency + w₂·bandwidth + w₃·cpu_load + w₄·throughput. Weights configured per traffic class in rule-based config. Normalized to 0–100 range.
Rule-based config maps traffic class (financial, bulk, real-time) to weight profile. Different classes can have entirely different optimization priorities on the same network.
Alternative path must score at least H points above current path to trigger switch. Prevents route flapping from small score oscillations. H configured per traffic class.
Switch triggers when current path score drops below threshold — before traffic quality degrades. Not triggered by path failure. Eliminates the "degradation window" before traditional failover.
The rule-based configuration layer abstracts routing decisions from low-level topology management. Operators express intent: "financial transaction traffic should always use paths with latency under 50ms; if no such path exists, alert operations and fall back to lowest-latency available." The optimization engine translates these rules into routing decisions as network conditions change, without requiring operators to manually update routing tables when conditions shift.
Rules support conditional logic, fallback chains, and alert thresholds — enabling sophisticated routing policy that would require constant manual intervention to maintain if managed at the routing protocol level. The rule engine is evaluated continuously as telemetry updates arrive, ensuring routing decisions always reflect current network state against current operator intent.
The system applies wherever mixed traffic classes share network infrastructure and static routing protocols leave performance on the table during congestion, partial degradation, or load imbalance events.
The independent claims cover the network route optimization system — the telemetry agent architecture, the path scoring engine, the rule-based configuration layer, and the proactive route switching mechanism. Dependent claims cover the four specific telemetry metrics (latency, bandwidth, CPU load, throughput), the weighted composite score formula, the hysteresis threshold mechanism, and the per-traffic-class rule configuration.
The claims are network-architecture agnostic — the invention applies to IP routing in any network topology (mesh, hub-spoke, fat-tree) and with any underlying protocol infrastructure. The key claim distinction from prior art is the combination of multi-metric continuous telemetry with rule-based traffic class optimization and proactive (not reactive) route switching — which together define a routing system that maintains performance commitments rather than simply recovering from failures.
US20260012411A1 is a pending application published in 2026. The application is currently under examination at the USPTO. Forward citations will be recorded after grant.
US20260012411A1 is the latest in a line of edge and distributed systems patents in the portfolio. US11611511B2 (Edge-Node Resource Distribution) established the foundational architecture for distributing computational resources across edge nodes based on real-time demand. This application extends that framework from resource distribution to route optimization — using the same multi-metric telemetry model to optimize data paths rather than compute allocation.
The rule-based configuration approach also connects to the security and authentication work in the portfolio: the same intent-based configuration model that governs transaction authorization rules (US11544718B2) here governs routing policy — enabling operators to express network performance intent in terms the optimization engine enforces automatically, rather than managing low-level protocol configuration.