PRT-DBA: A Predictive Real-Time Bandwidth Allocation Algorithm for Multi-Gigabit WANs

1. Introduction

The exponential growth of network traffic, driven by applications such as video streaming, cloud computing, and IoT, has placed significant demands on wide area networks (WANs). Multi-gigabit WANs, while offering high bandwidth, face challenges in dynamically allocating resources to meet the diverse QoS requirements of different traffic types. Traditional bandwidth allocation methods often lack real-time adaptability and predictive capabilities, leading to suboptimal performance under varying network conditions. This paper introduces the Predictive Real-Time Dynamic Bandwidth Allocation (PRT-DBA) algorithm, a novel approach designed to address these limitations. PRT-DBA combines predictive analytics with real-time adjustments to dynamically allocate bandwidth, ensuring optimal QoS for critical applications. The algorithm leverages machine learning models to predict future traffic patterns and continuously monitors network conditions to make real-time allocation decisions. By integrating predictive modelling and real-time monitoring, PRT-DBA provides a scalable and efficient solution for next-generation multi-gigabit WANs.

2. Related Work

The rapid growth of network traffic in multi-gigabit wide area networks (WANs) has driven significant research into dynamic bandwidth allocation (DBA) mechanisms that can adapt to real-time traffic conditions and predict future demands. Traditional approaches to bandwidth allocation often rely on static or heuristic methods, which are insufficient for handling the dynamic and unpredictable nature of modern network traffic. Recent advancements in machine learning (ML), predictive analytics, and Software-Defined Networking (SDN) have provided new opportunities to enhance network performance through real-time adaptability and predictive capabilities. This section reviews key contributions in the areas of predictive analytics, real-time monitoring, and dynamic bandwidth allocation, highlighting the gaps that the proposed Predictive Real-Time Dynamic Bandwidth Allocation (PRT-DBA) algorithm aims to address.

Predictive Analytics in Network Traffic Management: Predictive analytics has emerged as a powerful tool for network traffic management, enabling the forecasting of future traffic patterns and bandwidth demands. Early work by [1] proposed the use of time-series forecasting models, such as ARIMA (Auto Regressive Integrated Moving Average), to predict network traffic. While effective for short-term predictions, these models often struggle with capturing complex, non-linear traffic patterns. More recent research has explored the use of machine learning techniques, such as Long Short-Term Memory (LSTM) networks, for traffic prediction. For example, [2] demonstrated that LSTM-based models outperform traditional statistical methods in predicting traffic loads in data centre networks. Similarly, [3] proposed a hybrid approach combining LSTM and clustering techniques to improve the accuracy of traffic predictions in WANs.

Despite these advancements, existing predictive models often lack integration with real-time monitoring and dynamic bandwidth allocation mechanisms. This limits their ability to adapt to sudden changes in network conditions, such as traffic spikes or link failures. The proposed PRT-DBA algorithm addresses this limitation by integrating predictive analytics with real-time monitoring and dynamic allocation, ensuring optimal performance under varying network conditions.

Real-Time Monitoring and Dynamic Bandwidth Allocation: Real-time monitoring is critical for dynamic bandwidth allocation, as it enables the detection of anomalies and the adjustment of resource allocation in response to changing network conditions. Traditional monitoring approaches, such as those based on Simple Network Management Protocol (SNMP) and NetFlow, provide valuable insights into network performance but often lack the granularity and speed required for real-time adjustments. Recent research has explored the use of SDN and network function virtualisation (NFV) to enhance real-time monitoring capabilities. For instance, [4] proposed an SDN-based framework for real-time traffic monitoring and dynamic bandwidth allocation, demonstrating significant improvements in network performance and resource utilisation. Dynamic bandwidth allocation has also been a key focus in network management, particularly in the context of Quality of Service (QoS) and resource optimisation. Early work by [5] introduced dynamic allocation mechanisms for passive optical networks (PONs), which adjust bandwidth allocation based on traffic demand. More recent approaches have incorporated machine learning and predictive analytics to improve adaptability. For example, [6] proposed a reinforcement learning-based DBA algorithm for optical networks, demonstrating its effectiveness in reducing latency and improving throughput. However, these approaches often lack comprehensive mechanisms for real-time adjustments and fail to address the challenges of scalability and energy efficiency in multi-gigabit WANs.

Integration of Predictive Analytics and Real-Time Monitoring: The integration of predictive analytics and real-time monitoring has been a challenging research problem, as it requires balancing the accuracy of predictions with the speed of real-time adjustments. Recent advancements in machine learning and SDN have provided new opportunities to address this challenge. For example, [7] proposed an ML-based framework that combines traffic prediction with real-time monitoring to optimise bandwidth allocation in data centre networks. Similarly, [8] developed an SDN-based approach that uses predictive analytics to anticipate traffic spikes and adjust bandwidth allocation proactively. While these approaches show promise, they often lack scalability and fail to address the challenges of energy efficiency and security in multi-gigabit WANs.

Energy Efficiency and Security in Bandwidth Allocation: Energy efficiency has become a key concern in modern networks, particularly in large-scale WANs. Research by [9] explored energy-aware bandwidth allocation algorithms that minimize power consumption by consolidating traffic onto fewer network devices. Similarly, [10] proposed a dynamic bandwidth allocation scheme that reduces energy consumption by adjusting link rates based on traffic demand. However, these approaches often trade off energy efficiency for performance, highlighting the need for integrated solutions that balance these objectives. Security is another critical consideration in bandwidth allocation, as the increasing complexity of network traffic introduces new vulnerabilities. Research by [11] proposed a secure DBA framework that incorporates encryption and anomaly detection to protect against cyber threats. However, these approaches often lack integration with predictive analytics and real-time monitoring, limiting their effectiveness in dynamic network environments.

Research Gaps and Contributions: While significant progress has been made in predictive analytics, real-time monitoring, and dynamic bandwidth allocation, existing solutions often address these challenges in isolation. There is a lack of integrated frameworks that combine predictive modelling, real-time monitoring, and dynamic allocation to enhance network performance in multi-gigabit WANs. The proposed PRT-DBA algorithm bridges this gap by integrating these components into a unified framework. By leveraging machine learning models for traffic prediction, real-time monitoring for anomaly detection, and dynamic allocation for resource optimisation, PRT-DBA ensures optimal QoS, minimizes latency, and maximises throughput, making it a promising solution for next-generation networks.

3. Proposed Framework For PRT-DBA

3.1. Objectives

The primary objectives of PRT-DBA are:

• Predictive Traffic Modelling: Utilise machine learning models to predict future traffic patterns and bandwidth demands.
• Real-Time Monitoring: Continuously monitor network conditions to detect anomalies and adjust bandwidth allocation dynamically.
• Dynamic Bandwidth Allocation: Optimise bandwidth allocation to ensure QoS, minimise latency, and maximise throughput.
• Scalability: Operate efficiently in large-scale, high-capacity multi-gigabit networks.
• Energy Efficiency: Reduce energy consumption in network devices.

3.2. Key Features

i.

Predictive Traffic Modelling: PRT-DBA employs machine learning models, such as ARIMA and LSTM, to predict future traffic patterns based on historical data.

i.

Real-Time Monitoring: The algorithm continuously monitors network metrics, such as bandwidth utilisation, packet loss, and latency, to detect anomalies and adjust bandwidth allocation dynamically.

i.

Dynamic Bandwidth Allocation: PRT-DBA dynamically allocates bandwidth based on predicted and observed traffic conditions, ensuring optimal QoS for critical applications.

i.

Feedback Loop: The algorithm collects performance data to evaluate the effectiveness of allocation decisions and retrains predictive models to improve accuracy over time.

3.3. Algorithm Components

3.3.1. Data Collection and Pre-processing:

• Data Collection: Gather historical and real-time traffic data from network logs and monitoring tools.
• Pre-processing: Handle missing values, normalise traffic metrics, and extract relevant features.

3.3.2. Predictive Model Training

• Model Selection: Choose appropriate machine learning models for traffic prediction.
• Training and Validation: Train models using historical data and validate their accuracy.

3.3.3. Real-Time Traffic Monitoring

• Monitoring Setup: Configure network probes and monitoring tools to collect real-time metrics.
• Anomaly Detection: Detect anomalies or sudden changes in traffic patterns.

3.3.4. Bandwidth Allocation Engine

• Allocation Logic: Compute optimal bandwidth allocations based on predictions and real-time data.
• Optimization Techniques: Apply linear programming or heuristic algorithms to allocate bandwidth efficiently.

3.4. Key Workflow Summary

1.

Initialise: Configure queues and train predictive model.

2.

Monitor: Capture real-time traffic and compute stats.

3.

Predict: Forecast demand using ML model.

4.

Allocate Proactively: Assign bandwidth based on predictions.

5.

React to Errors:

a.

Increase Real-Time bandwidth if under-provisioned.

b.

Decrease Bulk bandwidth to compensate.

2.

Learn: Update model with new data for future cycles.

3.

Repeat: Iterate at fixed intervals.

Figure 1. Algorithm Flow.

Figure 1. Algorithm Flow.

This algorithm combines predictive intelligence with real-time reactivity, optimizing bandwidth for dynamic networks (e.g., 5G backhaul, cloud DCs) while self-correcting forecast errors.

Below is the PRT DBA Algorithm:

Algorithm 4. Predictive Real-Time Dynamic Bandwidth Allocation (PRT-DBA).

1.	function PRT-DBA(node, historicalData):
2.	// Initialization configureQueues()
3.	predictiveModel = trainPredictiveModel(historicalData)
4.	while simulationRunning:
5.	// Step 1: Real-Time Traffic Capture currentTraffic = captureTraffic(node)
6.	currentStats = analyzeCurrentStats(currentTraffic)
7.	// Step 2: Predict Future Demands predictedTraffic = predictTraffic(predictiveModel)
8.	// Step 3: Allocate Initial Bandwidth for each trafficClass in predictedTraffic
9.	initialAllocation = allocateBasedOnPrediction(trafficClass)
10.	setBandwidthAllocation(trafficClass, initialAllocation)
11.	// Step 4: Real-Time Adjustment if significantDeviationDetected(currentStats, predictedTraffic):
12.	for each trafficClass in currentStats:
13.	if trafficClass == "Real-Time": adjustBandwidth(trafficClass, increase)
14.	else if trafficClass == "Bulk": adjustBandwidth(trafficClass, decrease)
15.	// Step 5: Feedback Loop updatePredictiveModel(currentStats)
16.	wait(timeInterval)

3.5. PRT-DBA Algorithm Description

Line 1.	● Purpose: Main function for predictive real-time bandwidth allocation. Parameters:   i `node`: Network device (e.g., OLT, router) managing bandwidth.   ii `historicalData`: Past traffic patterns used for training predictive models.
Line 2.	● Action: Sets up queues for traffic classes (e.g., Real-Time, Bulk). ● Configuration: Queue sizes, scheduling policies (e.g., priority queuing).
Line 3.	● Action: Trains ML model (e.g., LSTM, Prophet) on historical traffic data. ● Output: Model capable of forecasting near-future traffic demands.
Line 4.	● Loop: Continuously executes bandwidth allocation during operation.
Line 5.	● Action: Monitors live traffic at the node (e.g., packets/bytes per second).
Line 6.	● Action: Computes real-time metrics: ● Per-class throughput, latency, queue occupancy. ● Identifies immediate congestion/starvation.
Line 7.	● Action: Forecasts traffic demand for next interval (e.g., 1–5 seconds). ● Input: Combines historical patterns + current traffic snapshots.
Line 8.	● Loop: Processes each traffic class for proactive allocation.
Line 9.	● Action: Assigns bandwidth based on predicted demand: ● E.g., Reserve 60% for Real-Time if surge is forecasted.
Line 10.	● Action: Enforces allocation on network hardware (e.g., via QoS policies).
Line 11.	● Condition: Triggers if actual traffic deviates from predictions (e.g., >20% error). ● Detection: Compares `currentStats` vs. `predictedTraffic` per class.
Line 12.	● Loop: Re-evaluates each class for corrective adjustments.
Line 13.	● Action: Boosts bandwidth for Real-Time (e.g., VoIP/video) if under-provisioned. ● Priority: Ensures SLA compliance during prediction errors.
Line 14.	● Action: Reduces bandwidth for Bulk traffic (e.g., file transfers). ● Goal: Frees capacity for higher-priority classes during shortages.
Line 15.	● Action: Retrains model with latest traffic stats. ● Purpose: Improves future predictions via continuous learning (online adaptation).
Line 16.	● Action: Pauses until next scheduling cycle (e.g., 100 ms–1 s). ● Balance: Allows frequent adjustments without overwhelming CPU.

4. Research Methodology

The methodology establishes a rigorous, reproducible framework for evaluating next-generation DBA algorithms in multi-gigabit WAN environments. By integrating realistic traffic models, validated failure scenarios, and statistically robust analysis techniques, we enable direct comparison of resilience and performance enhancements against industry standards. The ns-3 implementation balances fidelity with practicality, providing actionable insights for real-world deployment while transparently acknowledging scalability constraints.

4.2. Research Philosophy

• Pragmatic Paradigm: Combines quantitative simulation data with qualitative engineering insights to address real-world WAN challenges.
• Design Science Research (DSR): Focuses on designing, developing, and validating four novel DBA algorithms to optimize resilience and QoS.

4.3. Visual Workflow

Figure 2 shows the visual workflow:

5.2. Simulation Setup

5.2.1. Baseline Algorithms for Comparison

• Reactive DBA (R-DBA): Traditional threshold-based allocation.
• Machine Learning DBA (LSTM-DBA): Uses LSTM for traffic prediction.
• Proportional QoS-Aware (PQ-DBA): Prioritizes QoS without prediction.

5.2.2. PRT-DBA Parameters

• Prediction Model: Hybrid ARIMA + LightGBM (trained on traffic history).
• Features: Historical bandwidth usage, time-of-day patterns, flow priority.
• Retraining Interval: Every 5 minutes (online learning).
• Reallocation Speed: 5 ms decision cycles.

5.2.3. Network Configuration Parameters

Parameter	Value
Total Bandwidth	40 Gbps (multi-gigabit WAN)
Traffic Types	VoIP, Video, IoT, Bursty Data
QoS Requirements	Latency (<30ms), Packet Loss (<0.1%)
Prediction Window	10–50 ms (PRT-DBA’s look-ahead)
Congestion Scenarios	Periodic bursts, flash crowds

5.2.4. Key Performance Metrics

• Prediction Accuracy (MAE/RMSE for bandwidth demand).
• QoS Compliance (% of flows meeting latency/packet loss SLA).
• Overhead: Prediction time, algorithm complexity.
• Throughput Efficiency (Utilisation during congestion).

6.1. Prediction Accuracy

Table 1. Prediction Accuracy.

Algorithm	MAE (Mbps)	RMSE (Mbps)	Prediction Time (ms)
LSTM-DBA	12.5	18.2	8.2
PRT-DBA	6.8	9.1	3.5

Table 1. Prediction Accuracy.

Algorithm	MAE (Mbps)	RMSE (Mbps)	Prediction Time (ms)
LSTM-DBA	12.5	18.2	8.2
PRT-DBA	6.8	9.1	3.5

Figure 3. Prediction Accuracy.

Figure 3. Prediction Accuracy.

Finding: PRT-DBA reduces prediction error by 45% vs. LSTM-DBA, with 2× faster inference.

6.2. Latency Distribution (CDF)

Table 2. Latency Distribution.

Algorithm	90th Percentile Latency	99th Percentile Latency
R-DBA	45ms	80ms
LSTM-DBA	32ms	55ms
PRT-DBA	25ms	38ms

Table 2. Latency Distribution.

Algorithm	90th Percentile Latency	99th Percentile Latency
R-DBA	45ms	80ms
LSTM-DBA	32ms	55ms
PRT-DBA	25ms	38ms

Figure 4. Latency Distribution.

Figure 4. Latency Distribution.

6.3. Throughput Efficiency

Table 3. Throughput Efficiency.

Algorithm	Avg. Utilization	Utilization During Bursts
R-DBA	82%	70%
LSTM-DBA	88%	78%
PRT-DBA	95%	91%

Table 3. Throughput Efficiency.

Algorithm	Avg. Utilization	Utilization During Bursts
R-DBA	82%	70%
LSTM-DBA	88%	78%
PRT-DBA	95%	91%

Figure 5. Throughput Efficiency.

Figure 5. Throughput Efficiency.

Finding: PRT-DBA maintains 91% utilization during bursts (vs. 70% for R-DBA). C2. QoS Compliance (90% Load)

Table 4. QoS Compliance.

Metric	PRT-DBA	LSTM-DBA	R-DBA	PQ-DBA
Latency <30ms	98%	92%	85%	88%
Packet Loss <0.1%	99.5%	97%	90%	93%

Table 4. QoS Compliance.

Metric	PRT-DBA	LSTM-DBA	R-DBA	PQ-DBA
Latency <30ms	98%	92%	85%	88%
Packet Loss <0.1%	99.5%	97%	90%	93%

Figure 6. QoS Compliance.

Figure 6. QoS Compliance.

Findings: PRT-DBA achieves near-perfect SLA compliance even under flash crowds.

5. Discussion

5.1. Future Enhancements

Potential enhancements to PRT-DBA include:

• Advanced Machine Learning: Incorporate reinforcement learning for more adaptive bandwidth allocation.
• Emerging Technologies: Extend the algorithm to support 5G and IoT.
• SDN Integration: Integrate with SDN controllers for more flexible and programmable network management.

6. Conclusions

PRT-DBA represents a significant advancement in dynamic bandwidth allocation for multi-gigabit WANs. By integrating predictive analytics and real-time adjustments, the algorithm ensures optimal QoS, minimises latency, and maximises throughput. Simulation results demonstrate its effectiveness in improving network performance and resource utilisation. The proposed framework is scalable, secure, and compatible with existing network protocols, making it a promising solution for next-generation networks.