Peering Link Overflow Prediction for ISP Traffic Rerouting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data networks face challenges in predicting short-term data overflow situations, leading to traffic losses, degraded service quality, and increased costs due to insufficient interconnection capacities, especially during peak usage hours.
Innovation Solution
An automatic system using machine learning algorithms, specifically an ensemble of deep learning models like LSTNet and convolutional neural networks, predicts impending data overflows by monitoring traffic patterns and identifying specific interconnection nodes to reroute excess data, enabling proactive mitigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traffic is rerouted over alternative interconnection routes during overflow events, then service continuity is maintained, but the asymmetric capacities of alternative routes cause traffic to arrive at insufficient interconnection capacities, leading to degraded quality of service
Solution Approach 1:
The system performs preliminary actions by predicting overflow events before they occur using machine learning models that analyze historical traffic patterns. By identifying potential overflow situations in advance, the system can proactively reroute traffic through alternative interconnection routes before capacity exhaustion occurs, ensuring both service continuity and maintaining quality of service by avoiding the rush to reroute during actual overflow events
2Reliability
If interconnection link capacity is increased to handle peak traffic, then overflow events are prevented, but the cost of infrastructure expansion increases
Solution Approach 1:
The system implements dynamic traffic management by continuously monitoring traffic patterns and automatically rerouting traffic through alternative interconnection routes based on real-time conditions. This dynamic approach allows the network to adapt to varying traffic demands without requiring permanent infrastructure expansion, thereby preventing overflow events while avoiding the need to continuously maintain excess infrastructure capacity
Solution Approach 2:
The system changes operational parameters by adjusting traffic routing decisions based on predicted overflow conditions. Rather than changing the physical infrastructure capacity, the system modifies traffic flow parameters dynamically, redirecting traffic through different interconnection routes to balance load and prevent overflow, thus avoiding the need for infrastructure expansion
3Reliability
If traffic monitoring and prediction systems are implemented, then overflow events can be predicted and mitigated, but the complexity of the network management system increases
Solution Approach 1:
The system implements self-service by using machine learning models that automatically learn from historical traffic data and make autonomous predictions about future overflow events. The system self-adjusts routing decisions without requiring complex manual configuration or intervention, thereby achieving high prediction accuracy while keeping the management system relatively simple through automated, data-driven decision-making
Data Source
Figure 1~2
Figure 3~4
Figure 5a~6
AI summary
A system for early predicting of impending data overflow situations in data networks, comprising one or more sensors being networked computers that do not provide services, for collecting monitored data regarding network traffic volume from content providers to an Internet Service Provider (ISP) entering or exiting the ISP via peering autonomous systems connected to the ISP via physical links; one or more processors executing one or more deep learning models, being adapted to: identify over the data network being handovers alternative peering links, which are not inherent Private Network Interconnects (PNI) between the content providers and the ISP; determine static or dynamic threshold for dataflow anomaly associated with overflow; for each alternative handover, predict impending overflow situation by applying ML algorithms to the collected data that corresponds to the each alternative handover.