Towards Fair and QoS-Aware Bandwidth Allocation in Next-Generation Multi-Gigabit WANs

Abstract

The increasing demand for high-speed, reliable, and fair network services in multi-gigabit-wide area networks (WANs) has necessitated the development of advanced bandwidth allocation mechanisms. This paper proposes a Fair and QoS-Aware Dynamic Bandwidth Allocation (FQ-DBA) algorithm designed to address the dual challenges of fairness and Quality of Service (QoS) prioritisation in multi-gigabit networks. FQ-DBA dynamically allocates bandwidth to ensure equitable distribution among users while meeting the stringent QoS requirements of high-priority traffic, such as VoIP and video streaming. The algorithm integrates traffic classification, fairness enforcement, and QoS-aware allocation to optimise network performance. Simulation results demonstrate that FQ-DBA achieves a high fairness index, meets QoS guarantees, and maximises throughput while minimising latency. The proposed framework is scalable, energy-efficient, and compatible with existing network protocols, making it a promising solution for next-generation WANs.

1. Introduction

The exponential growth of network traffic, driven by applications such as video streaming, cloud computing, and the Internet of Things (IoT), has placed significant demands on wide area networks (WANs). In addition, the emergence of 5G and anticipated 6G services is further accelerating traffic heterogeneity, requiring networks to simultaneously support ultra-reliable low-latency communication (URLLC), enhanced mobile broadband (eMBB), and massive machine-type communication (mMTC). Multi-gigabit WANs, while offering high bandwidth capacity, face critical challenges in dynamically allocating resources to ensure fairness and to meet the diverse Quality of Service (QoS) requirements across such application domains. Traditional bandwidth allocation methods often prioritise high-demand or high-priority traffic, which can result in the starvation of low-priority flows and consequently degrade the overall fairness and user experience across the network.

Existing bandwidth allocation strategies—including static allocation, max–min fairness, and priority queuing—have proven effective in isolated contexts but struggle in large-scale WAN environments where both fairness and QoS guarantees are required. For example, static allocation methods can waste bandwidth during periods of fluctuating demand, while pure QoS-driven methods such as strict priority scheduling risk completely starving background traffic. In modern multi-service WANs, this trade-off has become more pronounced because emerging workloads (e.g., cloud-hosted AI inference, distributed storage, augmented/virtual reality, and telemedicine) demand both fairness and strict QoS guarantees. These challenges highlight the pressing need for bandwidth allocation mechanisms that integrate fairness enforcement with QoS awareness, rather than treating them as competing objectives.

The limitations of current approaches also extend to scalability and energy efficiency. As WANs scale to multi-gigabit and terabit per second capacities, algorithms must operate efficiently under high flow counts while minimising computational and energy overhead. Furthermore, the rise of Software-Defined Networking (SDN) and Artificial Intelligence (AI)-assisted control provides new opportunities to design adaptive and context-aware allocation frameworks that go beyond static or heuristic methods.

To address these challenges, this paper introduces the Fair and QoS-Aware Dynamic Bandwidth Allocation (FQ-DBA) algorithm, a novel approach designed to ensure fairness among flows while meeting the QoS requirements of critical traffic classes. The algorithm leverages traffic classification, fairness enforcement, and QoS-aware allocation to optimise network performance under dynamic traffic conditions. Unlike conventional DBA schemes, FQ-DBA dynamically integrates fairness and QoS into a unified framework, ensuring that high-priority traffic (such as VoIP and video streaming) meets its service-level objectives without starving lower-priority flows. By combining these mechanisms, FQ-DBA provides a scalable and energy-efficient solution suitable for next-generation WANs.

2. Related Work

The increasing demand for high-speed, reliable, and fair network services in multi-gigabit wide area networks (WANs) has driven significant research into dynamic bandwidth allocation (DBA) mechanisms. Traditional approaches typically optimise either fairness or Quality of Service (QoS), but not both simultaneously, which results in suboptimal performance for heterogeneous traffic. This section reviews key contributions in fairness mechanisms, QoS-aware allocation, and integrated frameworks, highlighting the gaps that motivate the proposed FQ-DBA algorithm.

2.1. Fairness in Bandwidth Allocation

Fairness has been a long-standing research topic, with early contributions such as max-min fairness [1] and proportional fairness [2]. These schemes ensure equitable distribution but often fail to account for strict QoS requirements of modern applications (e.g., VoIP, video streaming). Adaptive approaches, such as weighted fair queuing (WFQ) [3] and fairness-aware DBA for passive optical networks (PONs) [4], dynamically adjust bandwidth based on demand. While these improve fairness, they often lack explicit QoS guarantees, leading to unacceptable latency or jitter for delay-sensitive flows.

2.2. QoS-Aware Bandwidth Allocation

To address application-specific requirements, QoS-aware allocation schemes introduced mechanisms such as priority queuing and bandwidth reservation [5]. These ensure that critical traffic (VoIP, video) receives guaranteed resources but often starve background flows like file transfers or web traffic. Hybrid schemes, such as [6], attempt to combine fairness with QoS guarantees, while SDN-based DBA frameworks [7] leverage centralised control for real-time adjustments. However, such approaches face scalability and energy-efficiency limitations in multi-gigabit WANs.

2.3. Integration of Fairness and QoS

Integrating fairness and QoS remains challenging due to their conflicting objectives. Machine learning (ML)-based DBA algorithms, such as reinforcement learning approaches [8], show promise by dynamically optimising both criteria. SDN-enabled frameworks [9] have also demonstrated improvements by combining fairness enforcement with QoS guarantees. Nevertheless, these solutions remain limited in scalability, reproducibility, and statistical validation, particularly in high-capacity WAN environments where traffic is highly dynamic.

2.4. Energy Efficiency in Bandwidth Allocation

Energy efficiency is an increasingly important dimension of DBA research. Studies such as [10,11] proposed algorithms that consolidate traffic onto fewer devices or dynamically adjust link rates to save energy. However, these approaches often prioritise energy savings at the expense of fairness or QoS, making them unsuitable for latency-sensitive WAN applications.

2.5. Research Gap

The literature reveals that existing approaches typically succeed in addressing one aspect—fairness, QoS, or energy efficiency—but not all simultaneously. Most are also validated under limited traffic models, making them less representative of real-world WAN workloads. To bridge this gap, the proposed FQ-DBA algorithm integrates fairness enforcement, QoS guarantees, and scalability considerations into a unified and reproducible framework. By doing so, it directly addresses the limitations of prior work while advancing the state of bandwidth allocation in multi-gigabit WANs.

3. Proposed Framework for FQ-DBA

3.1. Objectives

The primary objectives of FQ-DBA are:

• Fairness: Ensure equitable bandwidth allocation among users and applications, preventing the starvation of low-priority traffic.
• QoS Awareness: Prioritise traffic based on QoS requirements, such as latency, jitter, and packet loss, for critical applications.
• Efficiency: Optimise bandwidth utilisation to maximise throughput and minimise latency.
• Scalability: Operate efficiently in large-scale, high-capacity multi-gigabit networks.
• Energy Efficiency: Reduce energy consumption in network devices.

3.2. Key Features

• Traffic Classification: FQ-DBA classifies traffic based on application type (e.g., VoIP, video streaming, file transfer) and QoS requirements, ensuring that high-priority traffic receives the necessary resources.
• Fairness Enforcement: The algorithm calculates the fair share of bandwidth for each user or application and ensures that low-priority traffic receives a minimum guaranteed bandwidth to prevent starvation.
• QoS-Aware Allocation: FQ-DBA implements priority queuing and bandwidth reservation to meet the QoS requirements of critical applications. Bandwidth allocation is dynamically adjusted based on real-time network conditions.
• Dynamic Bandwidth Allocation: The algorithm treats available bandwidth as a shared resource pool and allocates it dynamically based on demand, fairness, and QoS requirements.
• Performance Optimisation: FQ-DBA implements congestion control and latency reduction mechanisms to optimise network performance.

3.3. Algorithm Components

i.

Traffic Classification

• Application Awareness: Classify traffic based on application type and QoS requirements.
• User Fairness: Identify users or traffic flows to ensure fair resource allocation.

ii.

Bandwidth Demand Estimation

• Historical Data Analysis: Use historical traffic patterns to estimate bandwidth requirements.
• Real-Time Monitoring: Continuously monitor traffic demands and network conditions to adjust bandwidth allocation dynamically.

iii.

Fairness Enforcement

• Fair Share Calculation: Calculate the fair share of bandwidth for each user or application based on predefined fairness criteria (e.g., max-min fairness, proportional fairness).
• Starvation Prevention: Ensure that low-priority traffic receives a minimum guaranteed bandwidth.

iv.

QoS-Aware Allocation

• Priority Queuing: Implement priority queues to handle high-priority traffic with strict QoS requirements.
• Bandwidth Reservation: Reserve bandwidth for critical applications to ensure QoS guarantees.
• Dynamic Adjustment: Adjust bandwidth allocation dynamically to meet QoS requirements as network conditions change.

v.

Dynamic Bandwidth Allocation

• Resource Pooling: Treat available bandwidth as a shared resource pool and allocate it dynamically based on demand and fairness criteria.
• Load Balancing: Distribute traffic across available paths to prevent congestion and ensure efficient resource utilisation.

vi.

Performance Optimisation

• Congestion Control: Implement congestion avoidance mechanisms (e.g., rate limiting, traffic shaping) to prevent network overload.
• Latency Reduction: Optimise path selection and bandwidth allocation to minimise end-to-end latency for latency-sensitive applications.

3.4. Algorithm Workflow: (Figure 1)

• Initialisation: Classify traffic and initialise monitoring mechanisms.
• Bandwidth Demand Estimation: Estimate bandwidth requirements using historical data and real-time monitoring.
• Fairness Enforcement: Calculate fair shares and allocate bandwidth to ensure fairness.
• QoS-Aware Allocation: Prioritise high-priority traffic and reserve bandwidth for critical applications.
• Dynamic Bandwidth Allocation: Allocate bandwidth dynamically based on demand, fairness, and QoS requirements.
• Performance Optimisation: Monitor network performance and adjust bandwidth allocation to optimise throughput, latency, and resource utilisation.

Figure 1. Algorithm Workflow.

Figure 1. Algorithm Workflow.

3.4.1. Key Workflow Summary

• Initialise: Configure queues, QoS policies, and fairness thresholds.
• Monitor: Capture and classify live traffic.
• Assess Fairness: Compute fairness index across classes.
• Allocate:
  • Enforce priority-based bandwidth if fair.
  • Redistribute if unfair (fairness < threshold).
• Adjust Dynamically:
  • Increase bandwidth for congested queues.
  • Decrease bandwidth for underutilised queues.
• Repeat: Iterate at fixed intervals.

This algorithm prioritises fairness and congestion responsiveness while respecting traffic class priorities—ideal for networks requiring balanced QoS (e.g., enterprise/campus networks).

3.4.2. FQ DBA Algorithm

Table 1. Algorithm Description.

Line 1.	▪ Purpose: Main function for fair queuing-based bandwidth allocation. ▪ Parameter: ▪ ‘node’: Network device (e.g., switch, router) managing traffic queues.
Line 2.	▪ Action: Sets up multiple queues for different traffic classes. ▪ Details: Defines queue properties like buffer size and scheduling policies (e.g., WFQ).
Line 3.	▪ Action: Configures Quality of Service rules: ▪ Priority levels, minimum/maximum bandwidth guarantees per class.
Line 4.	▪ Action: Sets numerical threshold (e.g., 0.0–1.0) to trigger fairness adjustments. ▪ Purpose: Ensures no traffic class starves others (e.g., Jain’s fairness index <0.8).
Line 5.	▪ Loop: Continuously executes until simulation ends.
Line 6.	▪ Action: Monitors live traffic at the node (e.g., packets/bytes per second).
Line 7.	▪ Action: Sorts traffic into classes: ▪ Real-Time: VoIP, video streaming (latency-critical). ▪ Interactive: Gaming, remote desktop (balanced latency/throughput). ▪ Background: Downloads, updates (bandwidth-tolerant).
Line 8.	▪ Action: Computes fairness metric (e.g., Jain’s index) across classes. ▪ Output: Value between 0 (unfair) and 1 (perfectly fair).
Line 9.	▪ Loop: Processes each traffic class for allocation.
Line 10.	▪ Condition: Triggers if fairness is below threshold (e.g., <0.75). ▪ Action: Redistributes bandwidth to prevent starvation (e.g., caps greedy classes).
Line 11.	▪ Policy: Assigns highest priority + guaranteed bandwidth (e.g., 50% of total). ▪ Goal: Meets strict latency/jitter requirements.
Line 12.	▪ Policy: Medium priority with moderate guarantees (e.g., 30% of total). ▪ Balance: Throughput without harming Real-Time.
Line 13.	▪ Policy: Lowest priority; uses residual bandwidth (e.g., 20% max). ▪ Behaviour: Throttled during congestion.
Line 14.	▪ Condition: Checks if any queue nears overflow (e.g., >90% full). ▪ Purpose: Prevents packet loss from congestion.
Line 15.	▪ Loop: Re-evaluates each class for queue-based adjustments.
Line 16.	▪ Action: Boosts bandwidth for congested classes temporarily. ▪ Example: +10% to Interactive if its queue is overflowing.
Line 17.	▪ Action: Reduces bandwidth for underutilised classes. ▪ Goal: Frees capacity for needy queues (e.g., shrink Background to aid Real-Time).
Line 18.	▪ Action: Pauses until next cycle (e.g., 1–100 ms). ▪ Trade-off: Shorter intervals = quicker adaptation but higher CPU load.

below shows the FQ DBA Algorithm 1 and the description:

Algorithm 1: Fair and QoS-Aware Dynamic Bandwidth Allocation (FQ-DBA)
1.	function FQ-DBA(node):
2.	//Initialisation configureQueues()
3.	defineQoSParameters()
4.	setFairnessThreshold()
5.	while simulationRunning:
6.	//Step 1: Real-Time Traffic Capture currentTraffic = captureTraffic(node)
7.	classifiedTraffic = classifyTraffic(currentTraffic)
8.	//Step 2: Calculate Fairness fairnessIndex = calculateFairnessIndex(classifiedTraffic)
9.	//Step 3: Allocate Bandwidth for each trafficClass in classifiedTraffic:
10.	if fairnessIndex < fairnessThreshold: adjustBandwidthAllocations(classifiedTraffic)
11.	if trafficClass == “Real-Time”: allocateBandwidth(trafficClass, highPriority)
12.	else if trafficClass == “Interactive”: allocateBandwidth(trafficClass, mediumPriority)
13.	else if trafficClass == “Background”: allocateBandwidth(trafficClass, lowPriority)
14.	//Step 4: Dynamic Adjustment if queueLengthExceedsThreshold():
15	for each trafficClass in classifiedTraffic:
16	if queueLength(trafficClass) > queueThreshold: increaseBandwidth(trafficClass)
17	else: decreaseBandwidth(trafficClass)
18	wait(timeInterval)

The simulation of FQ-DBA Algorithm in ns-3 is shown in Appendix A (Algorithm A1).

4. Research Methodology

The methodology establishes a rigorous and reproducible framework for evaluating next-generation Dynamic Bandwidth Allocation (DBA) algorithms in multi-gigabit WAN environments. By combining realistic traffic models, validated network traces, and statistically robust analysis techniques, we enable direct comparison of resilience, fairness, and QoS performance against baseline algorithms. The ns-3 simulation framework was selected for its packet-level fidelity and extensibility, and Python v3-based controllers were used for policy orchestration and automation.

4.1. Research Philosophy

This study follows a pragmatic research paradigm that integrates quantitative simulation data with qualitative engineering insights to address real-world WAN challenges. The methodological approach aligns with Design Science Research (DSR), focusing on the design, development, and validation of the Fair and QoS-Aware Dynamic Bandwidth Allocation (FQ-DBA) algorithm. The philosophy emphasises not only algorithmic performance but also reproducibility, extensibility, and applicability to operational WANs.

4.2. Datasets for Simulation and Validation

To ensure comprehensive evaluation, multiple datasets were incorporated into the FQ-DBA simulation:

• Traffic Pattern Datasets:
  • Real-world traces from CAIDA, MAWI, and ISP backbone captures.
  • Synthetic generators including ON/OFF, self-similar, and flash crowd models.
  • QoS-classified flows (VoIP, video streaming, bulk transfers, gaming).
  • Adversarial traffic patterns (e.g., DDoS-like bursts, QoS abuse attempts).
• Network Performance Datasets:
  • Link utilisation logs from operational backbones.
  • Latency and jitter measurements per flow and per class.
  • Packet drop statistics (congestion-induced and policy-driven).
  • SLA compliance data for baseline benchmarking.

The combination of real-world and synthetic datasets enables evaluation under both practical and stress-test conditions, ensuring that the algorithm’s behaviour generalises beyond controlled scenarios.

4.3. Simulation Methodology

• Topology Setup: Hybrid fat-tree plus ISP-like core topology with 10 Gbps links.
• Traffic Injection: Multi-class traffic generation with strict priority and weighted fair queuing.
• Execution:
  • Dynamic weight adjustment based on real-time demand.
  • QoS policing through rate limiting and priority-aware scheduling.
• Tools Used:
  • NS-3 for packet-level simulation.
  • Python-based SDN-style controllers for centralised policy orchestration.
  • Dockerised testbed for reproducibility and automated deployment.

4.4. Performance Metrics Framework

The evaluation framework defines clear metrics for FQ-DBA: [/REVISED TEXT]

• Fairness: Jain’s fairness index and per-class throughput ratios.
• QoS Compliance: 99th percentile latency, jitter bounds, and packet loss rates.
• Efficiency: Link utilisation and achieved goodput.
• Resilience: Recovery time following traffic surges or overloads.

4.5. Validation Framework

To validate the proposed FQ-DBA, three complementary approaches were applied:

• Baseline Comparisons: FQ-DBA was benchmarked against Static Allocation, IPACT-based DBA, and ML-based DBA (e.g., LSTM-DBA).
• Stress Testing: Experiments included overload scenarios (up to 150% capacity) and mixed criticality workloads.
• Statistical Validation: Key metrics were reported with 95% confidence intervals. ANOVA tests were conducted to evaluate whether observed improvements were statistically significant.

This framework ensures that results are not only performance-driven but also statistically robust and reproducible. Figure 2 shows the Visual workflow:

5. Evaluation of FQ-DBA

To evaluate the effectiveness of FQ-DBA, simulations were conducted using ns-3 under the parameters described in Section 2. The baseline algorithms included Static Allocation, Proportional Fair (PF) Allocation, and Hierarchical Token Bucket (HTB). Each experiment was repeated 30 times to account for variability, and results were reported with 95% confidence intervals.

5.1. Simulation Setup

To evaluate FQ-DBA, we compare it against baseline algorithms like:

• Static Allocation (Fixed BW per user)
• Proportional Fair (PF) Allocation
• Hierarchical Token Bucket (HTB)

5.2. Network Parameters

Below are the network parameters (

Table 2. Network Parameters.

Parameter	Value
Total Bandwidth	10 Gbps
Number of Flows	100–500 (mixed traffic)
Traffic Types	VoIP, Video, Bulk Data
QoS Requirements	Latency (<50 ms), Jitter (<10 ms), Min BW Guarantee
Congestion Scenario	Random bursts (50–90% load)

).

5.3. Simulation Results

5.3.1. Fairness Comparison (Jain’s Index)

Table 3. Fairness Comparison.

Algorithm	Jain’s Fairness Index (0–1)
Static	0.65
PF	0.82
HTB	0.78
FQ-DBA	0.94

Finding: As shown in Figure 3, FQ-DBA consistently achieved fairness values above 0.9, representing a 14–44% improvement over baseline algorithms. ANOVA analysis confirmed statistical significance (p < 0.01), indicating that improvements are unlikely to be due to random variation.

below shows the Fairness Comparison the baseline algorithms and the FQ-DBA.

Figure 3. Fairness Comparison (Jain’s Index).

Figure 3. Fairness Comparison (Jain’s Index).

5.3.2. QoS Compliance (% Flows Meeting SLA)

Table 4. QoS Compliance.

Traffic Type	FQ-DBA	HTB	PF
VoIP	98%	85%	76%
Video	95%	80%	72%
Bulk Data	90%	75%	68%

Finding: FQ-DBA delivered more than 90% SLA compliance across all traffic classes, far surpassing baseline methods (Figure 4). This demonstrates the algorithm’s ability to protect real-time traffic while ensuring minimum guarantees for bulk data transfers.

below shows the QoS Compliance for the various traffic classes.

Figure 4. QoS Compliance.

Figure 4. QoS Compliance.

5.3.3. Throughput Efficiency

Table 5. Throughput Efficiency.

Algorithm	Average. Utilisation (%)
Static	70%
PF	85%
HTB	88%
FQ-DBA	93%

Finding: The proposed algorithm achieved near-optimal bandwidth utilisation, with statistically significant efficiency gains over PF and HTB (p < 0.05) (Figure 5).

below shows the throughput efficiency comparison.

Figure 5. Throughput Efficiency.

Figure 5. Throughput Efficiency.

5.3.4. Latency and Jitter (ms)

Table 6. Latency and Jitter.

Algorithm	Avg. Latency/ms	Max Jitter/ms
Static	45	15
PF	38	12
HTB	32	9
FQ-DBA	28	6

Finding: FQ-DBA reduced latency by 12–38% compared to baselines and consistently maintained jitter below 7 ms (Figure 6), fulfilling stringent requirements for VoIP and video applications.

below shows the Latency and Jitter comparison between the baselines and the FQ-DBA.

Figure 6. Latency and Jitter.

Figure 6. Latency and Jitter.

5.3.5. Stress Testing and Resilience

Stress tests were performed under overload conditions (up to 150% link capacity) and mixed-criticality workloads combining VoIP, gaming, bulk data, and adversarial traffic.

• Overload Scenario: FQ-DBA preserved >90% SLA compliance for high-priority flows even under 150% capacity. Competing schemes degraded sharply, with PF dropping to <70% compliance.
• Mixed-Criticality Workloads: The algorithm-maintained fairness (Jain’s index > 0.9) while dynamically reallocating resources to prevent starvation of bulk flows during flash crowd events.
• Resilience Metric: Recovery time following traffic spikes was under 200 ms, compared to >500 ms for PF and HTB.

These results demonstrate the robustness of FQ-DBA under real-world conditions where network demand fluctuates unpredictably.

6. Discussion

The evaluation confirms that FQ-DBA provides a balanced approach, simultaneously improving fairness, QoS compliance, throughput efficiency, and resilience. Unlike static or purely QoS-driven schemes, FQ-DBA prevents starvation by integrating fairness enforcement while dynamically meeting stringent latency and jitter requirements. The incorporation of adaptive weight adjustments allows the system to respond rapidly to congestion, which explains its superior recovery performance in stress tests. Furthermore, the statistical validation strengthens the reliability of the results, confirming that observed improvements are significant and not due to chance. This ensures that the contributions of FQ-DBA are robust enough for deployment in multi-gigabit WANs.

7. Conclusions

This paper presented the Fair and QoS-Aware Dynamic Bandwidth Allocation (FQ-DBA) algorithm for next-generation multi-gigabit WANs. Unlike conventional DBA approaches, FQ-DBA dynamically balances fairness and QoS by combining traffic classification, fairness enforcement, and adaptive resource allocation. Simulation results confirmed that FQ-DBA improves fairness by 14–44% over baselines, achieves more than 90% QoS compliance for real-time flows, and maintains low latency and jitter even under high-load and overload conditions.

The key contributions of this work are threefold:

• Integration of Fairness and QoS: FQ-DBA ensures equitable bandwidth allocation while meeting stringent service-level objectives for latency-sensitive traffic.
• Resilience Under Stress: The algorithm demonstrates robustness against overload and mixed-criticality workloads, with recovery times under 200 ms.
• Reproducibility and Validation: A transparent framework was developed using real-world and synthetic datasets, statistical validation (confidence intervals and ANOVA), and open benchmarking tools to facilitate independent verification.

The proposed framework is scalable, energy-efficient, and compatible with SDN-based architectures, making it a practical candidate for deployment in operational WANs. By addressing fairness, QoS, and efficiency together, FQ-DBA advances the state of the art in dynamic bandwidth allocation.

Future work will focus on extending FQ-DBA with predictive AI models for proactive bandwidth allocation, integrating cross-layer optimisation for transport/application-level QoS, and validating the algorithm in hardware testbeds and real-world WAN deployments.