AI Traffic QoS Profiles Using Model-Aware Compute and Latency Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current QoS configuration mechanisms for AI traffic do not account for the characteristic information of AI traffic, leading to inadequate satisfaction of AI traffic requirements and poor user experience.
Innovation Solution
Generate a QoS configuration for AI traffic flows that includes information about the AI model's computing power, latency, and user experience requirements, and transmit this configuration to communication apparatuses to enhance QoS management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current QoS configuration mechanisms are used for AI traffic, then general traffic transmission is supported, but AI traffic specific requirements (computing power, latency) are not adequately satisfied
Solution Approach 1:
The patent applies local quality by introducing AI-specific QoS parameters (computing power requirements, latency requirements, model indices) into the QoS configuration. This allows different parts of the QoS configuration to have different properties - general traffic parameters for standard flows and AI-specific parameters for AI traffic flows, thereby satisfying AI traffic requirements while maintaining general compatibility
Solution Approach 2:
The patent changes the parameters of QoS configuration by adding new AI-specific parameters such as computing power requirements, latency requirements, and AI model indices. This transforms the QoS configuration from a generic structure to one that can accommodate AI traffic characteristics, resolving the contradiction between reliability for AI traffic and adaptability to different traffic types
2Reliability
If QoS configuration includes detailed AI model information, then AI traffic requirements are better satisfied, but configuration complexity increases
Solution Approach 1:
The patent uses copying by introducing reference information about AI models (model indices, partitioning points) into the QoS configuration instead of duplicating entire model definitions. This allows the configuration to reference existing AI model characteristics without repeating all details, thereby maintaining reliability while reducing configuration complexity
Solution Approach 2:
The patent applies segmentation by dividing the QoS configuration into modular components - basic QoS parameters and AI-specific parameters (computing power, latency, model indices). This modular structure allows the system to handle AI traffic requirements reliably while keeping the configuration manageable through clear separation of concerns
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
This application relates to a communication method and apparatus, and a computer-readable storage medium. According to the technical solutions provided herein, a core network device generates a quality of service QoS configuration of a traffic flow, and sends the QoS configuration to a communication apparatus. The traffic flow is associated with an AI model. The QoS configuration includes information associated with the AI model. The information indicates at least one of the following: a requirement, in terms of computing power, of training or inference of the AI model; or a requirement, in terms of a latency, of training or inference of the AI model. Then, the communication apparatus may perform transmission of the traffic flow based on the QoS configuration. In this way, characteristic information of AI traffic can be fully taken into account to perform QoS configuration, and transmission of an AI traffic flow can be performed based on the QoS configuration, so that a requirement of the AI traffic can be more adequately satisfied and user experience of the AI traffic can be improved.