Hash-Based Preemption for Core Router Congestion Management

The patent extracts the complex per-flow state tracking and per-ingress-egress monitoring functions from the core network devices and relocates them to edge devices. The core router only maintains simple hash-bucket states and performs basic load measurements, while edge devices handle flow identification, state tracking, and congestion estimation. This extraction resolves the contradiction by maintaining preemption accuracy at the edge while simplifying core device complexity.

The patent introduces hash-buckets as intermediary aggregation units that group multiple flows together. Instead of tracking individual flows through the core, the system uses hash-buckets as mediators that aggregate flow states. The core router only needs to manage hash-bucket states rather than individual flow states, significantly reducing complexity while maintaining the ability to perform targeted preemption through the intermediary hash-bucket structure.

The patent extracts the policing function from core network devices and relocates it to edge devices. Edge devices are responsible for identifying flows, determining which flows to preempt based on hash-bucket states, and executing the preemption actions. This extraction makes policing easier at the edge where flow context is available, while core devices only perform simple load measurement and hash-bucket state management.

The patent implements a self-service mechanism where edge devices autonomously perform flow identification, congestion estimation, and preemption decisions based on hash-bucket states received from the core. The system does not require complex centralized control or intricate coordination between devices. Each edge device independently manages its own flows and preemption actions, simplifying overall system operation while maintaining QoS guarantees for non-preempted flows.

The patent applies local quality by performing precise per-flow congestion estimation at the edge devices where flow context is available, while using aggregated hash-bucket measurements only for high-level load monitoring at the core. Each edge device maintains accurate knowledge of individual flow states and performs precise congestion estimation locally, compensating for the aggregation at the core. This local precision resolves the contradiction by maintaining measurement accuracy where needed while accepting aggregation elsewhere.

The patent resolves the precision-loss issue by adding a spatial dimension to the measurement architecture. Instead of a single centralized measurement point, the system implements a hierarchical measurement structure with hash-bucket aggregation at the core (coarse-grained) and per-flow measurement at the edge (fine-grained). This multi-dimensional measurement approach allows the system to use aggregated data for overall load management while maintaining precise per-flow measurements at the edge for accurate congestion estimation and preemption decisions.