Random Early Detection Patent Landscape for Network Equipment

Overview of Technical Issues:

When network packet queues approach capacity, the packet dropping mechanism produces a harmful effect by blocking multiple incoming packets simultaneously in bursts, causing synchronized retransmission from all affected sources and resulting in network throughput oscillation and congestion collapse; the goal is to maintain stable queue levels and fair bandwidth distribution by providing early, probabilistic congestion signals before queue overflow occurs.

Solution directions generated for this problem

Problem Direction 1 :

ImproveCongestion signal timing
VS
ConstraintPacket processing overhead

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
This patent improves duration of action (extends fuel cell lifetime through early intervention) while avoiding loss of time by using [pre-established] comparison baselines and threshold devices that enable fast detection without complex real-time computation, directly paralleling the need for early congestion signaling with minimal per-packet processing overhead.
Fuel cell stack performance monitoring
Innovative Solution Refine solution

Pre-computed queue zone lookup table with hardware-accelerated drop decision

Pre-compute drop probability lookup table offline for queue zones 60-100%
How to solve :
  • Offline pre-calculate drop probability lookup table mapping 40 queue occupancy zones (60-100% in 1% increments) to drop probabilities (0-50% exponential curve)
  • store in on-chip SRAM as 40-entry array requiring only 160 bytes
  • each entry contains 8-bit drop threshold (0-255 scale) for hardware comparison
  • Implement hardware queue depth counter using dedicated register that auto-increments on enqueue and decrements on dequeue, updating current zone index (0-39) via simple bit-shift operation taking <0.1μs
  • counter feeds directly into table lookup circuit
  • At packet arrival, generate 8-bit hardware random number using linear-feedback shift register (LFSR) clocked at packet rate, compare against table threshold in single clock cycle (<0.2μs)
  • drop if random value exceeds threshold, total decision path under 0.8μs including queue insertion
Expected Effect : Per-packet processing <0.8μs; early signal at 60% capacity; smooth 0-50% drop curve
Risk Control :
  • SRAM access latency variation under high temperature
  • LFSR randomness quality degradation
  • queue counter synchronization errors during burst traffic

Inspiration 2 : Technology in this field

Search: Active Queue Management (AQM), Virtual Queue Marking, Early Congestion Detection, Queue Depth Monitoring, Low-Latency Signaling
Existing SolutionRefine solution

Random Early Detection with Instantaneous Queue Depth Monitoring

Use instantaneous queue depth to trigger probabilistic packet marking without averaging calculations
How to solve :
  • Implement instantaneous queue depth measurement by reading queue occupancy counter directly when packet arrives, eliminating exponential weighted moving average computation
  • Define four non-overlapping queue depth ranges (e.g., 0-60%, 60-75%, 75-90%, 90-100%) each mapped to distinct drop probability via pre-configured comparator thresholds stored in registers
  • Generate 3-bit random value using parallel LFSRs (7-bit, 8-bit, 9-bit) seeded per packet, compare against threshold to determine marking in single clock cycle
Expected Effect : Per-packet processing under 0.5μs; Early signaling at 60% capacity; Zero packet loss below threshold
Risk Control :
  • LFSR seed quality affecting randomness uniformity
  • Threshold calibration for different traffic patterns
  • Hardware register access latency

Problem Direction 2 :

ImproveDrop decision granularity
VS
ConstraintPacket processing overhead

Inspiration 1 : Cross-domain reference

Application Principle: #28 Mechanics substitution
Cross-domain applicability Assess applicability
This patent replaces time-consuming multi-test mechanical calibration with [hardware sensor] real-time measurement, improving positioning precision (measurement accuracy) while avoiding tool downtime (loss of time). It demonstrates how substituting computational/mechanical processes with dedicated hardware sensing enables precise continuous adjustment without processing delays, directly echoing the current contradiction of achieving smooth drop gradation precision while maintaining sub-microsecond packet processing speed.
Method and apparatus for measuring erosion and calibrating position for a mobile process kit
Innovative Solution Refine solution

Hardware LFSR-based probabilistic drop engine with pre-computed threshold registers

Replace software probability computation with hardware LFSR random generator and threshold comparator circuit
How to solve :
  • Implement 16-bit Linear Feedback Shift Register (LFSR) in hardware generating pseudo-random values 0-65535 every clock cycle (clocked at packet arrival)
  • Pre-load eight threshold registers mapping queue occupancy zones (60-65%: threshold=655 for 1% drop, 65-70%: 3277 for 5% drop, 70-75%: 6554 for 10%, 75-80%: 13107 for 20%, 80-85%: 19661 for 30%, 85-90%: 26214 for 40%, 90-100%: 32768 for 50%)
  • Packet forwarding logic reads current queue occupancy pointer, fetches corresponding threshold from register array via 3-bit zone index, compares LFSR output against threshold in single clock cycle—drop if LFSR value below threshold, forward otherwise
  • Background monitor updates zone pointer every 100 packets based on queue depth counter, decoupling monitoring from per-packet path
Expected Effect : Per-packet decision time <0.8μs; smooth drop curve 1-50%; hardware cost <500 logic gates
Risk Control :
  • LFSR periodicity causing pattern bias
  • threshold register update race condition
  • zone boundary hysteresis needed

Inspiration 2 : Technology in this field

Search: Random Early Detection (RED), Linear drop probability function, Hardware-accelerated probability calculation, Exponential drop function, PIE control scheme
Existing SolutionRefine solution

Hardware-Accelerated Exponential RED with Binary Shift Operations

Replace linear RED with exponential drop probability using hardware-optimized computation
How to solve :
  • Implement exponential drop probability DP = e^(-K*(maxTH-Avg)/(maxTH-minTH)) where minTH=60% queue, maxTH=100% queue, aggressiveness factor K=3-10 provides smooth 1-50% drop curve
  • minTH acts as early warning threshold triggering probabilistic drops before congestion
  • Replace division operations with binary right-shift for average queue calculation Avg = Avg_old*(1-2^-n) + Q_current*(2^-n) where n=9-11, completing in <100ns
  • use lookup table with 256 pre-computed e^-x values indexed by 8-bit queue position
  • Deploy bit-mask random number generator producing comparison value r in single clock cycle
  • drop packet if r ≤ DP, ensuring per-packet decision completes within 500ns including queue length sampling
Expected Effect : Sub-1μs processing; smooth 1-50% drop gradient; reduced retransmission synchronization
Risk Control :
  • Lookup table memory overhead
  • exponential parameter tuning across traffic patterns
  • maintaining drop uniformity during microbursts

Problem Direction 3 :

ImproveCongestion signal timing
VS
ConstraintNormal-operation packet loss rate

Inspiration 1 : Cross-domain reference

Application Principle: #3 Local quality
Cross-domain applicability Assess applicability
This patent addresses communication state transitions by applying [local quality] through differentiated control policies at different operational states, improving responsiveness (duration of action) while preventing unnecessary resource consumption (loss of substance), directly paralleling the need for selective congestion signaling that activates only at specific queue thresholds without causing packet loss during normal operation.
Communication state transitioning control
Innovative Solution Refine solution

Traffic-class-differentiated early congestion signaling with zero-loss baseline protection

Classify traffic into bulk and real-time classes using packet header inspection
How to solve :
  • Implement dual-policy queue management: classify incoming packets into bulk-transfer class (TCP file transfers, backups) and latency-sensitive class (VoIP, video conferencing, DNS) using port numbers and protocol flags in hardware packet classifier
  • Apply early probabilistic dropping only to bulk-transfer class starting at 60% queue occupancy with linear probability ramp (60%→1% drop, 70%→10% drop, 80%→30% drop) using pre-computed 5-entry lookup table indexed by queue depth register
  • Exempt latency-sensitive class from early dropping until 95% capacity, maintaining 0% loss baseline for real-time traffic during normal operation while bulk flows receive early congestion feedback
Expected Effect : Real-time traffic 0% loss; bulk traffic early signal at 65% queue; processing overhead <0.8μs
Risk Control :
  • misclassification of traffic types
  • lookup table indexing errors
  • queue occupancy measurement lag

Inspiration 2 : Technology in this field

Search: Active Queue Management (AQM), Random Early Detection (RED), Explicit Congestion Notification (ECN), Early Congestion Feedback, Packet Marking Probability
Existing SolutionRefine solution

Active Flow Random Early Drop with Per-Service Threshold Differentiation

Apply per-flow queuing within single FIFO buffer using flow identifiers (VLAN ID, IP ToS, service ID per IEEE 802.1ah); assign each flow unique minimum threshold minth and maximum threshold maxth based on QoS class (QCI+ARP), where premium flows get minth=70-80% and budget flows get minth=60-65% of buffer capacity to enable differentiated early warning
How to solve :
  • Implement mark-front strategy starting ECN marking when instantaneous queue length Q exceeds flow-specific minth (60-80% range), marking packets already in buffer from queue head rather than tail, delivering congestion notification 1 RTT faster than mark-tail
  • calculate drop probability Pa=maxp(Qavg−minth)/(maxth−minth) only when Q>maxth, with maxp dynamically adjusted per flow burst characteristics (increase maxp for bursty flows to maintain high link utilization)
Expected Effect : Apply <strong>packet drop prevention counter</strong> mechanism: after marking one packet at minth breach, accept next n packets (n=20-30 based on expected TCP window size) regardless of queue state to avoid multiple drops from same window, preventing global synchronization; reset counter when queue drains below minth ensuring zero packet loss during non-congested periods
Risk Control :
  • Congestion signaling starts at 60-80% capacity
  • zero packet loss below minth
  • fair bandwidth allocation across service classes
  • mark-front reduces feedback delay by 1 RTT

Problem Direction 4 :

ImproveQueue state stability
VS
ConstraintNormal-operation packet loss rate

Inspiration 1 : Cross-domain reference

Application Principle: #23 Feedback
Cross-domain applicability Assess applicability
This patent improves energy efficiency and component stability by using hardware-based feedback to transition current from pull-in to hold levels, preventing excessive stress. It resolves the contradiction between maintaining stable operation (improving stability) while reducing unnecessary energy loss (preventing substance loss), directly paralleling the queue management challenge of maintaining stability while minimizing packet drops.
Systems, methods, and devices for constant current relay control of a battery module
Innovative Solution Refine solution

Hardware-based dual-mode queue stability controller with adaptive feedback transition

Hardware feedback controller with dual-mode operation
How to solve :
  • Implement hardware-based queue monitor circuit with dual operating modes: aggressive mode (60-80% occupancy, 3-5% drop probability) activates only when queue variance exceeds 12% over 500ms sliding window
  • conservative mode (0.1% drop probability) engages when variance falls below 8% for 3 consecutive seconds, automatically transitioning between modes via hardware state machine without software intervention
  • Deploy dedicated FPGA logic block with 16-bit queue depth counter sampled every 50 packets, computing real-time variance using shift-register-based moving average (window size 20 samples), triggering mode transition via comparator circuit with 8% and 12% threshold registers, processing latency under 0.3μs per packet
  • Integrate hardware random number generator (linear feedback shift register, 32-bit) producing drop decisions by comparing LFSR output against mode-dependent threshold loaded from dual-port SRAM lookup table indexed by current queue occupancy percentage, maintaining per-packet processing under 0.8μs in both modes
Expected Effect : Queue variance reduced to ±5%; packet loss <0.8% during non-congestion; processing latency 0.8μs
Risk Control :
  • variance calculation overflow in burst traffic
  • mode transition hysteresis tuning
  • LFSR periodicity affecting randomness

Inspiration 2 : Technology in this field

Search: Active Queue Management (AQM), Random Early Detection (RED), Drop-Biasing Technique, Queue Occupancy Control, Adaptive Packet Dropping
Existing SolutionRefine solution

Periodic Early Discard (PED) with Fixed-Interval Drop Timer and Dual-Threshold Gating

Replace probabilistic dropping with fixed-period drop timer (τD) that triggers packet inspection at deterministic intervals, eliminating inter-drop time variance that causes global synchronization
How to solve :
  • Implement dual-threshold gating mechanism: drop next packet only when instantaneous queue length (IQL) exceeds minimum threshold bmin (60-64% of buffer) AND average queue length (AQL) exceeds gating threshold bgate (bmin/2) at timer expiration
  • accept all packets between timer events to prevent burst drops
  • Update drop period τD dynamically every 500ms based on AQL evolution: increase τD when AQL < bmin (multiplicative by factor including previous τD values to account for RTT delay in source reaction), decrease τD when AQL > bmax (80% buffer), reset to minimum (100μs) on buffer overflow, maintaining τD range 100μs-500ms to track phantom drop rate φ(t) that balances TCP source growth
Expected Effect : Queue occupancy stabilized at 60-80% range; packet loss <1% during non-congestion; 100% link utilization maintained; buffer requirement reduced to 3% of traditional sizing
Risk Control :
  • Accurate RTT estimation for control loop timing
  • coordination of τD adjustment speed with aggregate TCP time constants
  • emergency threshold tuning for fast congestion response

Problem Direction 5 :

ImproveCongestion signal timing
VS
ConstraintMust not deteriorate

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
This patent improves event prediction timing by identifying leading indicators that provide early warnings, allowing proactive measures before critical events occur. This directly parallels the current contradiction of extending congestion signal duration (early at 60-80% capacity) while preventing packet loss during normal operation, achieved through preliminary monitoring and threshold-based early action.
Identifying leading indicators for target event prediction
Innovative Solution Refine solution

Pre-indexed queue state lookup table with offline-computed drop thresholds for zero-latency congestion signaling

Pre-compute congestion response offline during initialization
How to solve :
  • Build a 256-entry lookup table during router initialization mapping queue depth percentages (0-100%) to drop probabilities (0-50%), stored in L1 cache for sub-nanosecond access
  • each entry pre-calculated using exponential curve P=0 for depth<60%, P=0.01×e^(0.08×(depth-60)) for 60-100%
  • Use hardware queue counter as direct index—packet arrival reads table[queue_depth_register] without computation, maintaining processing time at 0.5μs
  • Implement dual-phase activation logic: monitor 100ms rolling window of queue growth rate
  • enable table-based dropping only when growth exceeds 8%/second AND absolute depth >60%, otherwise force table index to zero-drop zone, achieving 0% loss during stable operation while providing early signals during congestion buildup
Expected Effect : Processing time ≤0.6μs; early signal at 62-65% capacity; 0% loss when growth rate <5%/second
Risk Control :
  • cache coherency in multi-core routers
  • table granularity insufficient for traffic bursts
  • growth rate calculation overhead

Inspiration 2 : Technology in this field

Search: Early congestion detection, Temporal congestion signals, Queue-based congestion control, Congestion notification signaling, Packet loss prevention
Existing SolutionRefine solution

Random Early Detection (RED) with Adaptive Threshold Configuration for Proactive Congestion Signaling

Deploy RED algorithm at network nodes to provide early congestion signals through probabilistic packet marking when average queue occupancy reaches configurable thresholds
How to solve :
  • Configure minimum threshold (MIN_th) at 60% queue capacity and maximum threshold (MAX_th) at 80% capacity
  • packets experience zero drop probability below MIN_th, linearly increasing drop/mark probability from 0% to Max_P (typically 10-15%) between MIN_th and MAX_th, and 100% marking above MAX_th
  • implement exponentially weighted moving average (EWMA) with weight factor 0.002-0.01 to calculate average queue size, filtering transient bursts: avg_queue = (1-weight) × prev_avg + weight × current_queue
  • mark packets using Explicit Congestion Notification (ECN) bits in IP header rather than dropping, allowing TCP sources to detect congestion via marked ACKs and reduce sending rates multiplicatively (typically halving congestion window) while increasing additively during non-congestion, achieving fairness across competing flows
Expected Effect : Queue occupancy stabilized at 60-75% range; packet loss reduced to <1% during congestion; throughput maintained at 95-98% link utilization
Risk Control :
  • Parameter sensitivity to traffic patterns requiring per-deployment tuning
  • EWMA weight selection balancing responsiveness versus stability
  • compatibility requiring ECN support across all network elements and end hosts
Patsnap Eureka Solution