Random Early Detection vs Weighted RED for QoS Networks
Overview of Technical Issues:
In standard Random Early Detection, the packet dropping decision module applies uniform random dropping that treats all traffic flows equally regardless of priority, creating a harmful effect where high-priority packets are dropped at the same rate as low-priority ones during congestion, directly violating QoS service level guarantees for critical applications; the goal is to implement priority-aware congestion management that protects high-priority flows while managing network congestion effectively.
Solution directions generated for this problem
Problem Direction 1 :
ImproveDrop probability differentiation capability
VSConstraintPacket classification complexity
Inspiration 1 : Cross-domain reference
Application Principle: #1 Segmentation
Cross-domain applicability
This patent improves adaptability by building flexible decision trees that handle diverse rule sets through [segmentation] and node merging, while preventing device complexity deterioration by optimizing tree structure and reducing memory requirements. It demonstrates how dividing classification tasks into optimized segments maintains versatility without increasing system complexity, directly addressing the current contradiction.
Decision tree level merging
Innovative Solution Refine solution
Two-stage binary priority filter with coarse-fine drop probability assignment
Split classification into two independent binary stages to minimize decision complexity
How to solve :
- Stage 1 binary filter: single bit check separates critical (DSCP ≥ 46) vs. non-critical traffic at ingress, O(1) operation using 6-bit DSCP field extraction only
- Stage 2 drop assignment: within critical group apply drop probability 0.01-0.05 (target <1% loss), within non-critical group apply 0.15-0.30 (15-30% loss acceptable), each stage uses pre-calculated lookup table indexed by single bit
- Implementation flow: packet arrives → extract DSCP bits 0-5 → compare bit pattern against threshold 46 (binary 101110) → set priority flag (1 bit) → queue occupancy triggers RED → read flag (O(1)) → index into 2-entry drop table → apply probability
Expected Effect : Classification reduced to 2 binary checks; QoS compliance >95% for critical traffic; per-packet overhead remains O(1); high-priority loss <1% vs. current 15-30%
Risk Control :
- DSCP field pre-marking dependency at upstream routers
- threshold calibration sensitivity to traffic mix changes
- lookup table synchronization under dynamic load
Inspiration 2 : Technology in this field
Search: Multi-level drop precedence differentiation, Priority-based queue management, Simplified packet classification, Proportional loss rate control, Active queue management
Existing SolutionRefine solution
Rate-Based Multi-Level Active Queue Management with Integral Control and Drop Precedence Differentiation
Use rate-based integral controller to match aggregate TCP connection rates to available capacity while maintaining minimal queue size
How to solve :
- Implement stable integral controller with gain ki calculated via closed-loop TCP/AQM plant model (ki range: -τkd0²z1√(z1²+d0²/τ²) < ki < 0, where z1∈(0,π/2) solves cot(z)=τz/d0)
- measure cumulative data arrival rate Rc(n) per precedence grade c as sum of current grade plus all higher-priority grades, compute normalized error e(n)=(T(n)-Rc(n))/x where T(n) is target capacity and x is nominal packet size
- calculate per-class drop probability pd,c(n)=pd,c(n-1)+kiΔte(n)/(2T(n)) where differentiation factor σc=1/pc based on service price pc, ensuring proportional loss distribution pd,c(n)=σc/Σσj·pd(n) across 3-4 priority classes
- apply exponentially weighted moving average filter Rc'(n)=(1-β)Rc'(n-1)+βRc(n) with β=0.002 for rate smoothing
- execute packet drop routine: if queue q<TL (no-drop threshold, TL=0.9T), accept packet
- else generate random pr∈[0,1], drop if pr≤pd,c(n)
- set control interval Δt and upper bound θ≤1 on drop probability
Expected Effect : High-priority traffic receives lowest discard rate; congestion control without short-term burst reaction; <1% high-priority packet loss during sustained overload
Risk Control :
- Integral controller gain stability under varying RTT distributions
- filter parameter tuning for burst tolerance versus responsiveness trade-off
- per-class threshold calibration accuracy
Problem Direction 2 :
ImproveQoS guarantee precision
VSConstraintDecision processing overhead
Inspiration 1 : Cross-domain reference
Application Principle: #10 Preliminary action
Cross-domain applicability
This patent improves measurement precision of QoS flow configuration by [pre-determining] supported bearers and flows during setup phase, avoiding loss of time during runtime data forwarding. It resolves the contradiction between accurate QoS classification and processing latency by performing bearer determination as a [preliminary action] before actual data transmission, directly echoing the current need to classify priority in advance to eliminate real-time inspection overhead.
Methods, apparatus and systems for integrated access and backhaul bearer management
Innovative Solution Refine solution
Pre-ingress priority tagging with metadata-driven O(1) drop decision engine
Extract priority once at ingress, store in metadata for O(1) drop lookup
How to solve :
- At packet ingress, extract 6-bit DSCP/ToS field from IP header and write to packet metadata tag (single memory write, ~5 CPU cycles)
- build 64-entry drop probability lookup table indexed 0-63, pre-populated with priority-class mappings (e.g., entries 0-15=80% drop for best-effort, 46-47=0.5% drop for EF class, 48-55=2% drop for AF4x)
- during congestion, RED engine reads metadata tag value as direct array index, retrieves drop probability in O(1) time without re-parsing headers
Expected Effect : >97% high-priority protection; <10ns decision latency; zero header re-inspection
Risk Control :
- metadata tag corruption in transit
- lookup table misconfiguration
- DSCP remarking by intermediate nodes
Problem Direction 3 :
ImprovePriority-based protection effectiveness
VSConstraintPacket classification complexity
Inspiration 1 : Cross-domain reference
Application Principle: #1 Segmentation
Cross-domain applicability
This patent applies [Segmentation] by dividing traffic into weighted groups managed by a central controller, improving Reliability (reducing packet loss through prioritized congestion control) while avoiding Device complexity (switches only enforce simple per-group throttle rates rather than complex per-flow decisions), directly matching the current contradiction of protecting high-priority packets without adding classification complexity.
Congestion control in packet data networking
Innovative Solution Refine solution
Binary priority flag with pre-ingress marking for minimal-complexity congestion control
Divide traffic into two segments at ingress
How to solve :
- At network ingress, mark high-priority packets with a single-bit priority flag in packet metadata (e.g., set bit 7 in internal header) based on DSCP values ≥46 or pre-configured application ports
- during congestion, RED drop decision reads only this 1-bit flag via O(1) memory access—apply drop probability 0.5% for flagged packets, 25% for unflagged packets when queue occupancy exceeds 70%
- implement ingress classification module using 64-entry lookup table indexed by DSCP field, storing binary output (0=best-effort, 1=critical), with table update latency <10μs via control plane API
Expected Effect : High-priority loss <1%, classification overhead <50ns per packet, no multi-tier logic
Risk Control :
- ingress marking inconsistency across devices
- flag bit position conflicts with existing metadata
- drop probability calibration under varying traffic patterns
Inspiration 2 : Technology in this field
Search: Weighted Random Early Detection (WRED), Priority-based packet marking, Selective packet dropping, Priority trie classification, Queue management with priority weighting
Existing SolutionRefine solution
Priority-Weighted Random Early Detection with Dynamic Threshold Adjustment
Assign priority-based flow weights to packets using application type, subscriber attributes, or service class identifiers at ingress, storing weights in per-flow state tables for O(1) lookup complexity
How to solve :
- Implement weighted drop probability calculation where p(i)=p(max)×(1-flow_weight)×(weighted_avg-Min)/(Max-Min), setting flow_weight=0.9 for high-priority traffic (VoIP, critical signaling) and flow_weight=0.2 for low-priority traffic (bulk data), ensuring high-priority packets receive 80% lower drop probability than low-priority packets at identical queue depths
- Configure dual-queue architecture with priority queue (Min=20%, Max=80% buffer) for high-priority flows and non-priority queue for best-effort traffic, applying immediate transmission scheduling (scheduled_time=current_time+1) when priority queue has packets, otherwise transmitting from non-priority queue to utilize guaranteed bandwidth fully
- Deploy congestion-aware priority elevation by monitoring per-flow packet loss rates in sliding windows of K packets (K=25-100), temporarily elevating priority of starved flows when loss exceeds threshold while maintaining aggregate conformance to traffic profiles, preventing complete starvation of lower-priority flows
Expected Effect : High-priority packet loss reduced from 15-30% to <1%; overall throughput maintained at link capacity; classification overhead O(1) per packet
Risk Control :
- Flow state table memory scaling with connection count
- priority inversion risk during extreme congestion
- backward compatibility with legacy TCP congestion control
