Dynamic VNF Resource Allocation via AI Traffic Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network function virtualization (NFV) systems face challenges in efficiently adjusting computing resource allocation to virtualized network functions (VNFs) due to fixed resource allocation methods, leading to potential waste or inability to respond to changing traffic patterns.
Innovation Solution
A method and server system utilizing artificial intelligence (AI) models, specifically a deep neural network (DNN) for predicting traffic and a multi-agent deep reinforcement learning model for adjusting computing resource allocation based on predicted traffic and status information from associated servers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fixed computing resources are allocated to VNFs, then resource allocation is simple and stable, but resource utilization efficiency deteriorates and the system cannot respond quickly to changing traffic patterns
Solution Approach 1:
The patent implements dynamic resource allocation by continuously monitoring traffic patterns and computing resource status, then adjusting VNF resource allocation in real-time based on predicted traffic demands and current system state, transforming the static fixed allocation into a dynamic adaptive system
Solution Approach 2:
The patent establishes a feedback loop where the orchestrator continuously receives status information from VNFs and computing resources, processes this information through AI models to predict future traffic, and adjusts resource allocation accordingly, creating a closed-loop control system that adapts to changing conditions
2Adaptability or versatility
If fixed computing resources are allocated to VNFs, then allocation management is straightforward, but the system loses adaptability to different traffic patterns and service requirements
Solution Approach 1:
The patent employs AI models to predict future traffic patterns and computing resource demands in advance, allowing the system to proactively adjust resource allocation before traffic spikes or drops occur, rather than merely reacting to changes
Solution Approach 2:
The patent dynamically changes allocation parameters such as CPU shares, memory allocation, and computing resource quotas for each VNF based on predicted traffic demands and current system utilization, allowing flexible adaptation to different service requirements
3Measurement precision
If AI models are used for traffic prediction and resource allocation, then prediction accuracy and resource optimization improve, but computational overhead and system complexity increase
Solution Approach 1:
The patent introduces an orchestrator as an intermediary layer between the VNFs and the management system, which houses the AI models and handles the complex computations for traffic prediction and resource allocation, isolating the complexity from the VNFs themselves
Data Source
AI summary
A method, performed by a server, of adjusting allocation of computing resources to a plurality of virtualized network functions (VNFs), and the server are provided. The method includes: for processing at least one task related to user equipments (UEs) connected to the server, identifying a plurality of VNFs related to the task; obtaining predicted traffic expected to be generated in the server by processing the task via the plurality of VNFs; obtaining, from at least one associated server, status information of computing resources in the at least one associated server; and adjusting allocation of computing resources to the plurality of VNFs based on the status information of the computing resources in the at least one associated server and the predicted traffic, wherein the at least one associated server includes another server that processes the task.