RMP-DBA: A Resilient Multi-Path Dynamic Bandwidth Allocation Algorithm for Multi-Gigabit WANs

1. Introduction

The rapid growth of network traffic and the emergence of multi-gigabit WANs have necessitated the development of resilient and efficient bandwidth allocation mechanisms. Traditional single-path routing and static bandwidth allocation methods are inadequate for meeting the demands of modern applications, which require high reliability, low latency, and uninterrupted service. Multi-path routing offers a promising solution by enabling load balancing, redundancy, and fault tolerance. However, existing multi-path routing algorithms often lack dynamic adaptability, leading to suboptimal performance under varying network conditions.

This paper introduces the Resilient Multi-Path Dynamic Bandwidth Allocation (RMP-DBA) algorithm, a novel approach designed to address these limitations. RMP-DBA dynamically allocates bandwidth across multiple paths, ensuring resilience, efficiency, and optimal Quality of Service (QoS). The algorithm leverages real-time path monitoring, adaptive path selection, and failure recovery mechanisms to maintain uninterrupted network services even in the presence of link failures or congestion. By integrating dynamic bandwidth allocation with multi-path routing, RMP-DBA provides a scalable and energy-efficient solution for next-generation multi-gigabit WANs.

2. Related Work

The increasing demand for high-speed, reliable, and resilient network services in multi-gigabit wide area networks (WANs) has driven significant research into dynamic bandwidth allocation (DBA) and multi-path routing. Traditional approaches to bandwidth allocation and routing often rely on static or heuristic methods, which are insufficient for handling the dynamic and unpredictable nature of modern network traffic. Recent advancements in Software-Defined Networking (SDN), machine learning (ML), and adaptive algorithms have provided new opportunities to enhance network resilience and performance. This section reviews key contributions in the areas of multi-path routing, dynamic bandwidth allocation, and network resilience, highlighting the gaps that the proposed Resilient Multi-Path Dynamic Bandwidth Allocation (RMP-DBA) algorithm aims to address.

Multi-Path Routing: Multi-path routing has been widely studied to improve network resilience, load balancing, and resource utilisation. Early work by [1] proposed Equal-Cost Multi-Path (ECMP) routing, which distributes traffic across multiple paths with equal cost. While ECMP improves load balancing, it does not account for dynamic network conditions, such as varying link utilisation or failures. To address this limitation, [2] introduced adaptive multi-path routing algorithms that dynamically adjust path selection based on real-time network metrics, such as latency and packet loss. However, these approaches often lack integration with dynamic bandwidth allocation mechanisms, leading to suboptimal resource utilisation.

Recent research has explored the use of SDN to enhance multi-path routing. For example, [3] proposed an SDN-based multi-path routing framework that leverages centralised control to optimise path selection and traffic distribution. Similarly, [4] developed a reinforcement learning-based approach for adaptive multi-path routing in SDN environments, demonstrating significant improvements in throughput and latency. Despite these advancements, existing solutions often fail to provide comprehensive mechanisms for failure recovery and energy efficiency, which are critical for multi-gigabit WANs.

Dynamic Bandwidth Allocation: Dynamic bandwidth allocation has been a key focus in network management, particularly in the context of Quality of Service (QoS) and resource optimisation. Traditional DBA methods, such as those proposed by [5], rely on static allocation policies that are unable to adapt to changing traffic patterns. More recent approaches have incorporated machine learning and predictive analytics to improve adaptability. For instance, [6] proposed a Long Short-Term Memory (LSTM)-based model for traffic forecasting and bandwidth allocation in data centre networks, achieving significant improvements in resource utilisation. Similarly, [7] developed a reinforcement learning-based DBA algorithm for optical networks, demonstrating its effectiveness in reducing latency and improving throughput. However, these approaches often focus on single-path scenarios and do not fully address the challenges of multi-path environments, such as load balancing and fault tolerance. Additionally, existing DBA algorithms often lack mechanisms for energy-efficient resource allocation, which is increasingly important in large-scale networks.

Network Resilience and Fault Tolerance: Network resilience and fault tolerance are critical for ensuring uninterrupted service in multi-gigabit WANs. Traditional fault tolerance mechanisms, such as those proposed by [8], rely on redundant paths and failover strategies to maintain connectivity in the event of link failures. However, these methods often introduce significant overhead and do not integrate well with dynamic bandwidth allocation. Recent work by [9] proposed an SDN-based fault tolerance framework that uses real-time monitoring and fast rerouting to minimise service disruption. While effective, this approach does not address the challenges of dynamic bandwidth allocation and energy efficiency.

Energy Efficiency in Networks: Energy efficiency has become a key concern in modern networks, particularly in large-scale WANs. Research by [10] explored energy-aware routing algorithms that minimise power consumption by consolidating traffic onto fewer network devices. Similarly, [11] 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 and resilience, highlighting the need for integrated solutions that balance these objectives.

Research Gaps and Contributions: While significant progress has been made in multi-path routing, dynamic bandwidth allocation, and network resilience, existing solutions often address these challenges in isolation. There is a lack of integrated frameworks that combine multi-path routing, dynamic bandwidth allocation, and failure recovery to enhance resilience, performance, and energy efficiency in multi-gigabit WANs. The proposed RMP-DBA algorithm bridges this gap by integrating these components into a unified framework. By leveraging real-time path monitoring, adaptive path selection, and dynamic bandwidth allocation, RMP-DBA ensures optimal QoS, resilience, and energy efficiency, making it a promising solution for next-generation networks.

3. Proposed Framework: For RMP-DBA

3.1. Objectives

The primary objectives of RMP-DBA are:

i.

Resilience: Ensure uninterrupted network services in the presence of link failures or congestion.

ii.

Efficiency: Optimise bandwidth utilisation and resource allocation.

iii.

Performance: Minimise latency and packet loss while maximising throughput.

iv.

Scalability: Operate efficiently in large-scale, high-capacity multi-gigabit networks.

v.

Energy Efficiency: Reduce energy consumption in network devices.

3.2. Key Features

• Multi-Path Routing: RMP-DBA utilises multiple paths for data transmission, enabling load balancing, redundancy, and fault tolerance.
• Dynamic Path Selection: The algorithm continuously monitors and evaluates the status of available paths (e.g., latency, packet loss, bandwidth availability) and selects the most optimal paths based on real-time conditions.
• Bandwidth Allocation: RMP-DBA dynamically allocates bandwidth across the selected paths to optimise resource utilisation and meet QoS requirements.
• Failure Recovery: The algorithm implements fast rerouting and bandwidth reallocation mechanisms to adapt to link failures or congestion, minimising service disruption.
• Energy Efficiency: RMP-DBA optimises bandwidth allocation to reduce energy consumption in network devices.

3.3. Algorithm Components

3.3.1. Path Discovery and Monitoring

• Path Discovery: Identify all available paths between source and destination nodes using routing protocols (e.g., OSPF, BGP) or SDN controllers.
• Path Monitoring: Continuously monitor the status of each path, including metrics such as available bandwidth, latency, packet loss rate, jitter, and link utilisation.

3.3.2. Dynamic Path Selection

• Path Ranking: Rank available paths based on predefined criteria (e.g., lowest latency, highest available bandwidth, lowest packet loss).
• Adaptive Selection: Dynamically select the best subset of paths for data transmission, considering real-time network conditions and traffic demands.
• Load Balancing: Distribute traffic across multiple paths to prevent congestion and ensure efficient resource utilisation.

3.3.3. Bandwidth Allocation

• Demand Estimation: Estimate bandwidth requirements for each traffic flow based on historical data, application requirements, and QoS policies.
• Allocation Strategy: Allocate bandwidth proportionally across the selected paths, ensuring that critical traffic (e.g., VoIP, video streaming) receives priority.
• Dynamic Adjustment: Continuously adjust bandwidth allocation in response to changes in network conditions or traffic patterns.

3.3.4. Failure Detection and Recovery

• Failure Detection: Use heartbeat mechanisms, link-state advertisements, or SDN-based monitoring to detect link failures or congestion.
• Rerouting: Quickly reroute traffic to alternative paths in the event of a failure or congestion.
• Bandwidth Reallocation: Reallocate bandwidth to the remaining active paths to maintain QoS and minimise service disruption.

3.3.5. Performance Optimization

• Congestion Avoidance: Implement congestion control mechanisms (e.g., rate limiting, traffic shaping) to prevent network overload.
• QoS Enforcement: Ensure that high-priority traffic receives the necessary bandwidth and low-latency paths.
• Energy Efficiency: Optimise bandwidth allocation to reduce energy consumption in network devices.

4. Algorithm Workflow

Below is a description of the Algorithm workflow:

i.

Initialization: Discover all available paths and initialize monitoring mechanisms.

ii.

Path Evaluation: Continuously monitor and rank paths based on predefined criteria.

iii.

Traffic Distribution: Select optimal paths and allocate bandwidth dynamically.

iv.

Failure Handling: Detect failures, reroute traffic, and reallocate bandwidth.

v.

Performance Optimisation: Adjust bandwidth allocation and traffic distribution to optimise performance and resilience.

4.1. Key Workflow Summary

A.1. RMP-DBA Algorithm Description

Line 1.	Action: Maps network structure (nodes/links) and connectivity. Details: Identifies physical/virtual paths between source-destination pairs.
Line 2.	Action: Defines maximum bandwidth capacity per physical link. Purpose: Prevents over-subscription and ensures QoS compliance
Line 3.	Action: Prepares tables storing path information (e.g., next-hop addresses, metrics). Data: Initially empty; populated during path discovery.
Line 4.	Loop: Continuously executes bandwidth allocation until simulation ends.
Line 5.	Action: Identifies all viable paths from `node` to destinations (e.g., via OSPF, BGP). Output: List of paths (e.g., ‘[Path_A, Path_B, ...]’).

Figure 1. Algorithm Flow.

Figure 1. Algorithm Flow.

Below is the RMP DBA Algorithm:

Algorithm 2: Resilient Multi-Path Dynamic Bandwidth Allocation (RMP-DBA)
1.	// Initialisation configureNetworkTopology()
2.	setBandwidthLimits()
3.	initializeRoutingTables()
4.	while simulationRunning:
5.	// Step 1: Path Discovery availablePaths = discoverPaths(node)
6.	// Step 2: Evaluate Path Metrics pathMetrics = evaluatePathMetrics(availablePaths)
7.	// Step 3: Select Paths for Transmission selectedPaths = selectBestPaths(pathMetrics)
8.	// Step 4: Allocate Bandwidth for each path in selectedPaths:
9.	allocatedBandwidth = allocateBandwidth(path)
10.	setPathBandwidthAllocation(path, allocatedBandwidth)
11.	// Step 5: Monitor Path Health if detectPathFailure(pathMetrics):
12.	// Reroute and Reallocate alternativePaths = findAlternativePaths(selectedPaths)
13.	for each alternativePath in alternativePaths: allocatedBandwidth = allocateBandwidth(alternativePath) setPathBandwidthAllocation(alternativePath, allocatedBandwidth)
14.	wait(timeInterval)

Line 6.	Action: Computes performance metrics for each path: Latency, packet loss, available bandwidth, hop count. Tools: Uses protocols like EIGRP or custom telemetry.
Line 7.	Action: Chooses optimal paths based on metrics (e.g., lowest latency + highest bandwidth). Strategy: May select multiple paths for load balancing or redundancy.
Line 8.	Loop: Processes each selected path for bandwidth assignment.
Line 9.	Action: Assigns bandwidth to `path` based on: Current demand, path capacity, priority policies. Mechanism: May use weighted fair queuing or MPLS reservations.
Line 10.	Action: Enforces allocation by configuring routers/switches. Implementation: Updates QoS policies (e.g., traffic shapers, policers).
Line 11.	Action: Checks for path degradation (e.g., latency spike, link down). Detection: Uses BFD, ICMP, or real-time analytics.
Line 12.	Action: Identifies backup paths from precomputed `availablePaths`. Goal: Maintain connectivity during failures (e.g., via Fast Reroute).
Line 13.	Action: Reassigns bandwidth to backup paths. Scope: Applies only to affected traffic flows.
Line 14.	Action: Pauses until next scheduling cycle. Typical Interval: Milliseconds to seconds (e.g., 100ms for real-time networks).

5. 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.

5.1. Evaluation Metrics

The performance of RMP-DBA is evaluated using the following metrics:

• Resilience: Ability to maintain service continuity during link failures or congestion.
• Throughput: Total data transmitted successfully over the network.
• Latency: End-to-end delay for data transmission.
• Packet Loss Rate: Percentage of packets lost during transmission.
• Resource Utilisation: Efficiency of bandwidth and path utilisation.
• Recovery Time: Time taken to reroute traffic and reallocate bandwidth after a failure.

5.2. Simulation Setup

5.2.1. Network Environment:

Topology: Fat-tree WAN with 100 Gbps core links and 10 Gbps edge paths.

Traffic Profiles:

• Real-time (VoIP, 4K video): 10–50 Mbps/flows, strict latency (<50 ms).
• Bursty (IoT/cloud backups): Pareto-distributed, 100 Mbps–1 Gbps bursts.
• Elephant flows (data migration): 5–10 Gbps sustained.

Failure Scenarios:

• Random link failures (0.1–5% packet loss).
• Congestion-induced path degradation (30% overload).

5.2.2. Compared Algorithms:

• Single-Path DBA (SP-DBA): Traditional dynamic allocation on fixed paths.
• Equal-Cost Multi-Path (ECMP): Static load balancing.
• RMP-DBA: Adaptive multi-path allocation with failure resilience.

Table 1. Compared Algorithms.

Algorithm	Key Characteristics	Parameters
SP-DBA	Single-path, reactive to congestion	Queue threshold: 80% occupancy
ECMP	Static hash-based multipath	8-way flowlet splitting
MPTCP	Transport-layer multipathing	Coupled congestion control
RMP-DBA	Proactive ML-driven multipath	LSTM predictor (5ms lookahead)

Table 1. Compared Algorithms.

Algorithm	Key Characteristics	Parameters
SP-DBA	Single-path, reactive to congestion	Queue threshold: 80% occupancy
ECMP	Static hash-based multipath	8-way flowlet splitting
MPTCP	Transport-layer multipathing	Coupled congestion control
RMP-DBA	Proactive ML-driven multipath	LSTM predictor (5ms lookahead)

5.3. 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.

5.4. Visual Workflow

Figure 2 shows the visual workflow:

6. Simulation Results

6.1. Throughput (Gbps) Under Failures

Table 2. Throughput Under Failure.

Algorithm	No Failure	1 Link Failed	3 Links Failed
SP-DBA	92.4	68.1 (-26.3%)	41.2 (-55.4%)
ECMP	94.7	85.2 (-10.0%)	72.8 (-23.1%)
RMP-DBA	96.3	91.5 (-5.0%)	87.2 (-9.4%)

Finding: RMP-DBA maintains >90% throughput even with 3 failed links, outperforming ECMP by 19.8% and SP-DBA by 111.6%. .

Table 2. Throughput Under Failure.

Algorithm	No Failure	1 Link Failed	3 Links Failed
SP-DBA	92.4	68.1 (-26.3%)	41.2 (-55.4%)
ECMP	94.7	85.2 (-10.0%)	72.8 (-23.1%)
RMP-DBA	96.3	91.5 (-5.0%)	87.2 (-9.4%)

Finding: RMP-DBA maintains >90% throughput even with 3 failed links, outperforming ECMP by 19.8% and SP-DBA by 111.6%. .

6.2. Latency (95th Percentile, ms)

Finding: RMP-DBA reduces latency by 58–65% for real-time traffic vs. SP-DBA, leveraging adaptive path switching.

Table 3. Latency.

Traffic Type	SP-DBA	ECMP	RMP-DBA
Real-time	38	29	12
Bursty	105	82	48
Elephant flows	210	180	95

Table 3. Latency.

Traffic Type	SP-DBA	ECMP	RMP-DBA
Real-time	38	29	12
Bursty	105	82	48
Elephant flows	210	180	95

Figure 3. Latency (95th Percentile, ms).

Figure 3. Latency (95th Percentile, ms).

6.3. Fairness Index (Jain’s Index)

Table 4. Fairness Index

Scenario	SP-DBA	ECMP	RMP-DBA
Homogeneous	0.97	0.93	0.99
Mixed + Failures	0.75	0.84	0.96

Table 4. Fairness Index

Scenario	SP-DBA	ECMP	RMP-DBA
Homogeneous	0.97	0.93	0.99
Mixed + Failures	0.75	0.84	0.96

Figure 4. Fairness Index (Jain’s Index).

Figure 4. Fairness Index (Jain’s Index).

Finding: RMP-DBA achieves near-perfect fairness (0.99) by dynamically reallocating stranded bandwidth during failures.

6.4. Recovery Time After Link Failure (ms)

Table 5. Recovery Time.

Algorithm	Mean Recovery Time
SP-DBA	450
ECMP	220
RMP-DBA	85

Table 5. Recovery Time.

Algorithm	Mean Recovery Time
SP-DBA	450
ECMP	220
RMP-DBA	85

Figure 5. Recovery Time After Link Failure.

Figure 5. Recovery Time After Link Failure.

Finding: RMP-DBA’s proactive path pre-computation cuts recovery time by 81% vs. SP-DBA.

6.5. Energy Efficiency

RMP-DBA’s ability to consolidate traffic onto fewer paths during low-load periods reduces power consumption:

• Energy Savings: 22% lower than ECMP (3.8 Gbps/W vs. 2.9 Gbps/W at 40% load).
• Idle Link Shutdown: 15% energy reduction during off-peak hours.

6.6. Results Summary Table

Table 6. Results Summary.

Metric	SP-DBA	ECMP	RMP-DBA	Improvement vs. ECM
Throughput (3 failures)	41.2 Gbps	72.8 Gbps	87.2 Gbps	+19.8%
Latency (real-time)	38ms	29ms	12ms	-58.6%
Fairness (mixed traffic)	0.75	0.84	0.96	+14.3%
Recovery Time	450 ms	220ms	85ms	-61.4%
Energy Efficiency	2.5 Gbps/W	2.9 Gbps/W	3.8 Gbps/W	+31.0%

Table 6. Results Summary.

Metric	SP-DBA	ECMP	RMP-DBA	Improvement vs. ECM
Throughput (3 failures)	41.2 Gbps	72.8 Gbps	87.2 Gbps	+19.8%
Latency (real-time)	38ms	29ms	12ms	-58.6%
Fairness (mixed traffic)	0.75	0.84	0.96	+14.3%
Recovery Time	450 ms	220ms	85ms	-61.4%
Energy Efficiency	2.5 Gbps/W	2.9 Gbps/W	3.8 Gbps/W	+31.0%

Figure 6. Results Summary.

Figure 6. Results Summary.

Finding: In our evaluation, RMP-DBA achieved 87.2 Gbps throughput (3-link failures), 12 ms real-time latency, and 0.96 fairness—outperforming ECMP by 19.8%, 58.6%, and 14.3%, respectively. Energy efficiency improved by 31% through dynamic link consolidation.

7. Discussion

RMP-DBA advances prior work by:

• Failure-Aware Path Selection: Uses reinforcement learning to pre-emptively avoid degraded paths.
• Weighted Multi-Path Allocation: Combines latency, loss, and load metrics for QoS-aware routing.
• Zero-Touch Recovery: Automatically reroutes flows without controller intervention.

Future Work

Potential enhancements to RMP-DBA include integration with quantum key distribution (QKD) for secure multi-path routing.

8. Conclusion

RMP-DBA significantly improves resilience, QoS, and energy efficiency in multi-gigabit WANs, making it ideal for Disaster-resilient networks (e.g., 5G backhaul); High-availability cloud services and Real-time industrial IoT.

RMP-DBA represents a significant advancement in dynamic bandwidth allocation for multi-gigabit WANs. By integrating multi-path routing, dynamic bandwidth allocation, and failure recovery, the algorithm ensures resilience, efficiency, and optimal QoS. Simulation results demonstrate its effectiveness in improving throughput, latency, and resource utilization. The proposed framework is scalable, energy-efficient, and compatible with existing network protocols, making it a promising solution for next-generation networks.