ML-Based VNF Placement and Segment Routing Under Link Utilization Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining optimal placement of virtual network functions (VNFs) and routing in networks are inefficient, leading to suboptimal resource utilization, increased latency, and higher operational costs.
Innovation Solution
A machine learning model is used to jointly optimize VNF placement and routing by learning from traffic matrices and network topology, minimizing maximum link utilization while constraining VNF capacity and count, using backpropagation and gradient descent to determine optimal routing variables and placement locations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional methods are used for determining VNF placement and routing, then the system is simpler to implement, but resource utilization is suboptimal and operational costs are higher
Solution Approach 1:
The patent replaces traditional manual or rule-based VNF placement and routing methods with a machine learning model that automatically learns optimal configurations from traffic matrices and network topology, substituting mechanical decision-making processes with intelligent algorithms
Solution Approach 2:
The machine learning model learns optimal routing variables and VNF placement configurations by adjusting parameters based on traffic patterns and network conditions, dynamically changing system parameters to achieve optimal resource utilization
2Reliability
If suboptimal VNF placement and routing are used, then the system is easier to implement, but latency increases and performance decreases
Solution Approach 1:
The machine learning model is trained in advance on traffic matrices and network topology data to learn optimal VNF placement and routing strategies before actual network operations, enabling proactive optimization rather than reactive adjustments
Solution Approach 2:
The system uses learned routing variables and placement configurations as feedback to continuously optimize network performance, adjusting decisions based on observed traffic patterns and performance metrics
3Productivity
If optimal VNF placement and routing are implemented using machine learning, then resource efficiency is enhanced and costs are reduced, but the system requires more complex training and computation
Solution Approach 1:
The machine learning model performs computationally intensive training and learning operations in advance, during off-peak periods or offline, so that the actual network operations use pre-learned optimal configurations with minimal real-time computational overhead
Solution Approach 2:
The system discards suboptimal routing paths and VNF placements that do not meet performance criteria, focusing computational resources only on learning and maintaining the most efficient configurations
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
In some embodiments, there may be provided a method that includes learning, by a machine learning model and using at least the at least one traffic matrix and the information regarding the network topology of the network, a first output comprising a routing variable defined as a fraction of traffic that originates from the source node to the destination node via via an intermediate node and a second output comprising placement within the network of the at least one virtual network function; and outputting, by the machine learning model, the first output and the second output to configure segment routing on the network and to configure placement of the at least one virtual network function. Related systems and articles of manufacture are also described.