Compare Random Early Detection vs Weighted RED for QoS

Overview of Technical Issues:

In standard Random Early Detection, the packet queue management mechanism discards incoming packets randomly without differentiating by traffic priority class, creating a harmful effect where high-priority flows experience unnecessary packet loss equal to low-priority flows, violating quality of service guarantees; the goal is to implement priority-aware congestion control that protects critical traffic while managing network congestion effectively.

Solution directions generated for this problem

Problem Direction 1 :

ImproveTraffic differentiation capability
VS
ConstraintPacket processing complexity

Inspiration 1 : Cross-domain reference

Application Principle: #1 Segmentation
Cross-domain applicability Assess applicability
This patent applies [Segmentation] by dividing user interaction processing into profile creation (offline) and profile lookup (runtime) stages, improving system adaptability for personalized multi-modal services while preventing computational complexity from degrading real-time performance. This mirrors the current need to enhance traffic differentiation versatility without increasing per-packet processing complexity.
Multiple wake words for systems with multiple smart assistants
Innovative Solution Refine solution

Bitmap-indexed priority tagging with single-cycle drop decision lookup

Bitmap priority tagging at ingress with direct lookup drop decision
How to solve :
  • Encode 4 priority levels using 2-bit bitmap tags (00=critical, 01=high, 10=medium, 11=low) inserted into packet metadata during mandatory routing header inspection—zero additional processing stage required
  • Implement pre-computed 2D decision matrix [queue_depth_bin × priority_tag] stored in L1 cache (64-byte table: 16 queue bins × 4 priorities), mapping directly to binary drop/accept—single memory read replaces multi-step calculation
  • Update queue depth bins via hardware counter sampling every 100 packets (1% overhead), maintaining 8 depth ranges (0-12.5%, 12.5-25%...87.5-100%) with priority-specific thresholds: critical flows protected until 90% full, low-priority dropped starting at 40% full
Expected Effect : Processing overhead ≤2 operations per packet; high-priority loss <0.8%; throughput reduction <3%
Risk Control :
  • cache coherency in multi-core routers
  • bitmap tag collision with existing metadata fields
  • decision matrix calibration for varying traffic patterns

Inspiration 2 : Technology in this field

Search: Multi-level priority classification, Two-stage packet processing, Header field inspection, Priority-based queue management, Early action termination
Existing SolutionRefine solution

Weighted Random Early Detection with Priority-Based Threshold Differentiation

Implement priority-aware congestion control using Weighted RED with differentiated thresholds
How to solve :
  • Assign separate min_th and max_th thresholds for each of 3-4 priority classes (e.g., min_th_high=100, max_th_high=150 packets
  • min_th_low=50, max_th_low=100 packets) based on packet priority field inspection
  • Calculate single average queue length (avg_ql) shared across all classes, then apply class-specific drop probability P_drop = max_p × (avg_ql - min_th)/(max_th - min_th) where max_p varies by class (high-priority max_p=0.02, low-priority max_p=0.10)
  • Perform one priority field read plus one threshold comparison per packet to select appropriate parameter set, ensuring processing overhead remains ≤2 operations while higher-priority packets experience lower drop probability at same queue occupancy
Expected Effect : High-priority loss <1% vs. 5-15% in standard RED; 3-4 priority levels; ≤2 operations overhead
Risk Control :
  • Threshold calibration for traffic mix variations
  • Parameter tuning across priority classes
  • Backward compatibility with non-priority traffic

Problem Direction 2 :

ImproveDrop decision precision
VS
ConstraintPacket processing complexity

Inspiration 1 : Cross-domain reference

Application Principle: #28 Mechanics substitution
Cross-domain applicability Assess applicability
This patent replaces complex real-time computational systems (accelerometer data processing) with [embedded sensor-based detection] and pre-processed lookup mechanisms, improving measurement precision (accurate flight time detection) while avoiding device complexity increase (eliminating external equipment). It demonstrates how substituting mechanical/computational systems with field-based detection and pre-computed data structures resolves the contradiction between accuracy and processing overhead.
Flight time
Innovative Solution Refine solution

Pre-computed 3D drop decision tensor with single-cycle indexed retrieval

Replace runtime calculation with indexed tensor lookup
How to solve :
  • Pre-compute a 3D drop decision tensor [queue_depth × priority_class × random_seed] during router initialization, storing binary drop/accept decisions for all state combinations (e.g., 256 queue levels × 4 priority classes × 256 random values = 256KB lookup table)
  • Encode packet priority as 2-bit tag during ingress routing phase when headers are already inspected, eliminating redundant classification at queue
  • Execute drop decision via single tensor index operation: decision = tensor[current_queue_depth][packet_priority_tag][hardware_random_byte], reducing per-packet processing from 5 steps (classify + measure + calculate + compare + decide) to 1 memory access
Expected Effect : Processing overhead <2 operations per packet; high-priority loss <1%; throughput maintained at baseline
Risk Control :
  • tensor memory footprint exceeds L1 cache causing access latency
  • priority tag injection fails at ingress under extreme load
  • pre-computed probabilities drift from optimal under dynamic traffic patterns

Inspiration 2 : Technology in this field

Search: Class-based drop probability, Priority-based thresholds, Per-packet processing, Weighted RED algorithm, Class-specific thresholds
Existing SolutionRefine solution

Priority-Indexed Random Early Detection with Per-Class Threshold Mapping

Assign priority-indexed drop probabilities using class-specific thresholds with minimal computation overhead
How to solve :
  • Classify incoming packets into 3-4 priority classes using DSCP/ToS bits at ingress, mapping each to a pre-calculated drop probability curve with class-specific min_th and max_th parameters (e.g., high-priority: min_th=80%, max_th=95%
  • low-priority: min_th=20%, max_th=80%)
  • Implement single queue depth measurement shared across all classes, calculating average queue length once per sampling interval using exponentially weighted moving average with weight w=0.002, then apply class-specific thresholds via table lookup requiring only 1-2 additional operations (array index + comparison)
  • Configure differentiated max_p values per class (high-priority max_p=0.02, low-priority max_p=0.20) stored in pre-computed lookup tables indexed by [class][queue_depth_range], enabling O(1) drop probability retrieval without per-packet calculation
Expected Effect : High-priority loss <1%, low-priority adaptive dropping, <5% processing overhead
Risk Control :
  • Threshold calibration under varying traffic loads
  • Queue depth sampling accuracy with bursty traffic
  • Lookup table memory footprint in hardware implementations

Problem Direction 3 :

ImproveQoS guarantee reliability
VS
ConstraintPacket processing complexity

Inspiration 1 : Cross-domain reference

Application Principle: #3 Local quality
Cross-domain applicability Assess applicability
This patent improves prediction reliability by applying [local quality] through differentiated motion information processing paths—simple paths for close-value references and complex paths for diverse references—thereby enhancing performance without proportionally increasing decoding complexity, mirroring the current need to improve high-priority flow reliability while constraining per-packet processing overhead.
Bi-predictive image decoding method
Innovative Solution Refine solution

Dual-path queue architecture with fast-bypass for high-priority flows

Separate traffic into fast-bypass and standard paths based on priority
How to solve :
  • Implement dual-path queue architecture: high-priority packets (DSCP EF/AF41) enter dedicated express queue with simple tail-drop at 95% threshold, low-priority packets use shared queue with standard RED
  • ingress classifier reads pre-existing DSCP field (already in IP header) and routes to appropriate queue in single comparison operation, adding only 1 processing step
  • Express queue sized at 15-20% of total buffer capacity, services high-priority flows with strict priority scheduling, guarantees <1% loss under congestion by isolating from RED dropping mechanism
  • Shared queue maintains RED parameters (min_th=40%, max_th=80%, max_p=0.1) for aggregate congestion control, absorbs 80-85% of traffic, prevents starvation via weighted fair queuing with 80% minimum bandwidth allocation
Expected Effect : High-priority loss <1%, processing overhead +1 operation, throughput maintained at 98% baseline
Risk Control :
  • express queue starvation of low-priority flows
  • buffer allocation imbalance under traffic pattern shifts
  • DSCP field trust boundary enforcement

Inspiration 2 : Technology in this field

Search: Priority-based packet dropping, Queue depth-based congestion control, Active flow random early drop, Packet loss rate monitoring, Flow-selective preemption
Existing SolutionRefine solution

Weighted Random Early Detection with Priority-Based Threshold Differentiation

Assign differentiated priority classes using existing IP precedence or DSCP fields already present in packet headers, requiring only single field read operation
How to solve :
  • Implement priority-indexed threshold arrays where each class (P1/P2/P3) maps to pre-calculated min_th, max_th, and max_p values: P1 gets min_th=40%, max_th=80%, max_p=0.02
  • P2 gets min_th=30%, max_th=70%, max_p=0.10
  • P3 gets min_th=20%, max_th=60%, max_p=0.25, enabling single array lookup operation per packet
  • Calculate drop probability using standard RED formula Pa=max_p×(Q_avg−min_th)/(max_th−min_th) with class-specific parameters, where higher priority classes have larger min_th (delayed drop trigger), larger max_th (extended safe zone), and lower max_p (gentler drop rate), requiring only one additional comparison and multiplication beyond standard RED
  • Maintain per-flow drop tracking flag to prevent multiple drops within 100 RTT window for high-priority flows, using lightweight boolean check as second operation, ensuring P1 traffic experiences preferential protection during congestion while P3 absorbs majority of drops
Expected Effect : P1 packet loss reduced from 5-15% to <1%; P2 loss 2-5%; P3 loss 10-20%; processing overhead <2 operations per packet
Risk Control :
  • Threshold calibration for specific traffic mix
  • starvation prevention for low-priority flows requiring minimum bandwidth guarantee
  • interaction with TCP congestion control requiring coordination

Problem Direction 4 :

ImproveTraffic differentiation capability
VS
ConstraintSystem throughput capacity

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
This patent improves system adaptability (handling dynamic SRS requests accurately) while preventing productivity degradation (maintaining throughput) by [pre-determining] transmission rules and detecting control information in advance, ensuring efficient resource allocation without real-time processing overhead during critical periods—directly paralleling the need to enhance traffic differentiation capability while preserving forwarding rate.
Communication device and SRS transmission control method
Innovative Solution Refine solution

Ingress-stage priority tagging with cached drop decision tables for minimal-overhead QoS differentiation

Tag packets at ingress during routing lookup phase to avoid re-inspection at queue
How to solve :
  • Embed priority classification into existing ingress routing lookup stage where headers are already inspected — extract DSCP/ToS field and map to 2-bit priority tag (00=critical, 01=high, 10=normal, 11=low) stored in packet metadata, adding <1% processing overhead
  • Pre-compute drop probability lookup tables during router initialization mapping (queue_depth_percentage, priority_tag) to binary drop/accept decision — 4 priority levels × 100 queue depth steps = 400-entry table stored in L1 cache, enabling single-cycle lookup
  • At congestion point, RED engine reads pre-assigned 2-bit tag from packet metadata and performs single table lookup [queue_depth_index][priority_bits] instead of header inspection plus calculation, limiting per-packet processing to 1 memory access (3-5 CPU cycles) versus baseline 2 cycles — total overhead <3%
Expected Effect : Per-packet time +2.8%, critical traffic loss <0.8%, forwarding rate maintained at 99.7% baseline
Risk Control :
  • Cache coherency under multi-core forwarding
  • table size exceeding L1 cache capacity
  • priority tag corruption in metadata pipeline

Inspiration 2 : Technology in this field

Search: Multi-level packet priority classification, Two-stage packet processing, Priority-based queue management, Differentiated congestion control, Drop precedence mechanisms
Existing SolutionRefine solution

Load-Tolerant Multi-Precedence WRT Active Queue Management in a Single FIFO Queue

Use threshold-based precedence demotion in one FIFO queue so critical packets keep lower drop exposure without adding heavy per-packet work.
How to solve :
  • Implement WRT-style active queue management on a single FIFO queue with 4 drop precedences, maintaining cumulative average queue lengths per level and demoting packets to the next lower precedence when the associated threshold is exceeded
  • Configure shared max threshold with ordered demotion thresholds: one common max_th, one common no-drop th_in, and descending demotion thresholds for top classes, with RED/WRED min_th and max_p per precedence to preserve strict hierarchy under overload
  • Realize classification by fast header lookup and cached precedence bits using DSCP or service-ID mapping, fixed-point EWMA queue estimators, SRAM-resident tables, and periodic conformance tests comparing measured drop ratios, queue occupancy, and forwarding cycles against acceptance envelopes
Expected Effect : high-priority loss sharply reduced; strict priority hierarchy preserved under overload; forwarding overhead within near-baseline range; starvation risk minimized
Risk Control :
  • threshold misconfiguration under mixed TCP UDP traffic
  • classifier mapping consistency across ingress paths
  • overload transition stability and fairness tuning

Problem Direction 5 :

ImproveDrop decision precision
VS
ConstraintMust not deteriorate

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
This patent improves measurement precision of frame drop decisions by [pre-setting] differentiated thresholds for various frame types, enabling fine-grained control without adding runtime complexity. It resolves the contradiction between decision accuracy (measurement precision) and processing speed, directly matching the current need to achieve <1% high-priority loss precision while maintaining fast processing throughput.
Video processing methods, devices, servers and media
Innovative Solution Refine solution

Pre-computed priority-indexed drop decision table for zero-latency class-specific RED

Pre-compute all drop decisions offline and retrieve via single table lookup at runtime
How to solve :
  • During router initialization, pre-calculate a 3-dimensional drop decision table indexed by (queue_depth, priority_class, random_seed) covering all operational states — queue depth discretized into 64 levels (0-100% in 1.56% steps), 4 priority classes (0=critical to 3=best-effort), 256 random values, yielding 65,536 pre-computed binary drop decisions stored in 8KB L1-cache-resident lookup table
  • At packet arrival, perform single-cycle table lookup using current queue occupancy counter (maintained via lightweight increment/decrement on enqueue/dequeue), pre-assigned 2-bit priority tag from ingress classification, and hardware random number generator output — total processing reduced to 1 memory read + 3 register operations (≤5ns on modern NPU)
  • Configure class-specific thresholds offline: priority-0 min_th=70% max_th=95% max_p=0.008, priority-1 min_th=60% max_th=90% max_p=0.02, priority-2 min_th=40% max_th=80% max_p=0.08, priority-3 min_th=20% max_th=70% max_p=0.15, ensuring <1% loss for critical traffic while maintaining congestion control for lower classes
Expected Effect : High-priority loss <0.8%, throughput penalty <2%, lookup latency 5ns
Risk Control :
  • table size exceeds cache capacity
  • queue depth quantization introduces threshold boundary errors
  • random seed distribution bias affects drop uniformity

Inspiration 2 : Technology in this field

Search: Class-based packet drop probability, Priority-based traffic differentiation, Weighted random early detection, Burst threshold detection, Header-based fast classification
Existing SolutionRefine solution

Class-Based Bandwidth Partitioning with Per-Flow Drop Probability Control

Implement priority-aware packet admission control using measured arrival rates and fair share calculations to protect critical traffic
How to solve :
  • Establish per-class virtual queues with separate arrival rate tracking (Mi) and drop probability (Di) calculated as Di=Dmax*(1-flow_weight)*(avg_queue-Min)/(Max-Min), where flow_weight ranges 0-1 based on priority class
  • Update parameters periodically (every 10-20ms) in control path while data path performs fast per-packet decisions using pre-computed probabilities, checking if arrival rate ≥ burst threshold to drop immediately or comparing random number R against Di
  • Assign higher flow_weight (e.g., 0.9 for P1 traffic vs 0.2 for best-effort) to achieve differentiated drop probabilities, ensuring high-priority packets drop at only 10% rate of low-priority packets during congestion
Expected Effect : High-priority loss <1%, 5-10x lower drop probability than low-priority traffic
Risk Control :
  • Periodic parameter update frequency tuning
  • Burst threshold calibration for different traffic patterns
  • Fair share calculation accuracy under dynamic load
Patsnap Eureka Solution