Edge Cloud QoS Manager Using Neural Network Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multi-access edge computing (MEC) systems face challenges in dynamically managing quality of service (QoS) and resource allocation, particularly in IoT and 5G network communications, leading to inefficiencies in latency and bandwidth optimization.
Innovation Solution
The implementation of a MEC QoS manager that employs Top-down Microarchitecture Analysis Method (TMAM) metrics and process monitoring counters to identify bottlenecks and dynamically adjust computing resources, using prediction logic based on neural networks to anticipate and prevent performance degradation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If computing resources are dynamically adjusted in MEC systems, then resource allocation efficiency is improved, but system complexity increases
Solution Approach 1:
The patent implements prediction logic using neural networks that analyze historical traffic patterns and performance metrics to forecast future resource requirements before actual demand occurs. This preliminary action enables proactive resource allocation, adjusting computing resources in advance based on predicted traffic configurations, thereby improving allocation efficiency while reducing the need for complex real-time reaction mechanisms
Solution Approach 2:
The system employs continuous monitoring of performance metrics including latency, throughput, and resource utilization, feeding this data back to the neural network predictor and resource allocator. This closed-loop feedback mechanism enables dynamic adaptation where the system learns from actual performance and refines its resource allocation decisions, improving efficiency through data-driven optimization while managing complexity through iterative improvement rather than complex upfront design
2Reliability
If neural network prediction logic is implemented, then performance degradation is prevented, but computational overhead increases
Solution Approach 1:
The patent implements prediction logic that operates at strategic intervals rather than continuously, using neural networks to forecast resource requirements for upcoming time windows. The system performs partial prediction actions focused on critical performance parameters and key decision points, avoiding excessive continuous computation. This approach maintains performance stability by making accurate predictions when needed while reducing computational overhead by avoiding unnecessary continuous prediction operations
Solution Approach 2:
The neural network predictor performs computational work in advance to generate predictions about future traffic patterns and resource requirements. By completing prediction computations before actual resource allocation decisions are needed, the system ensures accurate predictions are available when required for maintaining performance stability, while spreading computational load over time rather than concentrating it during critical allocation moments
3Loss of time
If computing resources are moved closer to users in MEC, then latency is reduced, but network management complexity increases
Solution Approach 1:
The patent segments the network into distributed MEC hosting points positioned closer to end users, with each MEC node independently managing its own computing resources and local traffic. This segmentation reduces latency by enabling local processing without requiring all management functions to be centralized. Each MEC segment operates semi-autonomously with the neural network predictor and resource allocator deployed locally, reducing the need for complex inter-node coordination while maintaining low-latency local service delivery
Data Source
AI summary
A device of a service coordinating entity includes communications circuitry to communicate with a plurality of access networks via a corresponding plurality of network function virtualization (NFV) instances, processing circuitry, and a memory device. The processing circuitry is to perform operations to monitor stored performance metrics for the plurality of NFV instances. Each of the NFV instances is instantiated by a corresponding scheduler of a plurality of schedulers on a virtualization infrastructure of the service coordinating entity. A plurality of stored threshold metrics is retrieved, indicating a desired level for each of the plurality of performance metrics. A threshold condition is detected for at least one of the performance metrics for an NF V instance of the plurality of NFV instances, based on the retrieved plurality of threshold metrics. A hardware resource used by the NFV instance to communicate with an access network is adjusted based on the detected threshold condition.