Machine Learning Segment Routing for Changing Traffic Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing routing schemes in networks are inflexible and prone to congestion due to insufficient robustness in handling varying traffic patterns, such as time-of-day, seasonal, and event-driven demands.
Innovation Solution
A machine learning-based approach that transforms linear deflection parameters into non-linear parameters to optimize link utilization across a network, using gradient descent to adjust deflection parameters for efficient traffic distribution, thereby determining a robust routing scheme that minimizes maximum link utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional routing schemes are used, then network setup is simple, but the routing is inflexible and prone to congestion under varying traffic patterns
Solution Approach 1:
The patent implements dynamic routing by using machine learning models to continuously adapt route selection based on real-time traffic patterns. The system transitions from static routing to dynamic routing where paths are automatically adjusted according to observed traffic conditions, time-of-day variations, seasonal patterns, and event-driven demands, thereby achieving routing flexibility without manual intervention.
Solution Approach 2:
The patent changes the parameters of routing decisions by using machine learning models that process multiple input features (traffic matrices, time stamps, event indicators) to generate optimized routing parameters. The system transforms routing from fixed parameter-based decisions to adaptive parameter adjustments based on learned patterns, enabling flexible response to varying network conditions.
2Reliability
If traditional routing schemes are used, then implementation is simple, but network resilience is insufficient under varying traffic demands
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors traffic patterns, compares them against learned models, and adjusts routing decisions accordingly. The machine learning models process feedback from traffic matrices and temporal patterns to refine future routing decisions, enabling the network to adapt to varying demands and maintain resilience under different operational conditions.
Solution Approach 2:
The patent applies preliminary action by using machine learning models to predict future traffic patterns based on historical data and temporal patterns. The system proactively adjusts routing before congestion occurs by anticipating demand changes based on time-of-day patterns, seasonal variations, and event indicators, thereby enhancing network resilience before problems arise.
3Productivity
If machine learning-based routing optimization is implemented, then traffic distribution is optimized and congestion is reduced, but computational complexity increases
Solution Approach 1:
The patent segments the routing optimization problem by processing different aspects of traffic management separately. The machine learning model handles high-level routing decisions based on traffic matrices and temporal patterns, while the network infrastructure executes specific routing actions. This segmentation allows the complex optimization to be distributed across multiple levels, reducing the computational burden on any single component.
Solution Approach 2:
The patent uses machine learning models to create virtual copies of network behavior patterns from historical data. By learning from replicated traffic patterns and simulating future scenarios, the system can optimize routing decisions without requiring exhaustive computational exploration of all possible routes, thereby reducing computational complexity while maintaining optimization effectiveness.
Data Source
AI summary
In some embodiments, there may be provided a method that includes receiving a first traffic matrix; receiving information regarding links associated with each segment of the network; determining a total amount of segment flow using the at least one non-linear deflection parameter applied to the traffic demand of the first traffic matrix; determining a link flow for each of the links using the total amount of segment flow and the second input to the machine learning model; determining link utilization for each of the links using the link flows and a capacity for each of the links; learning, by the machine learning model using a gradient descent, a minimum of a maximum amount of the link utilization over the links by at least adjusting a value of the at least one non-linear deflection parameter. Related systems, methods, and articles of manufacture are also disclosed.