AI Packet Processing Architecture with Forwarding Plane Pre-Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network devices face inefficiencies when integrating artificial intelligence (AI) for processing network packets due to the high computational demands, particularly in complex network scenarios, where the forwarding plane's capabilities are often overwhelmed, leading to bottlenecks in data processing.
Innovation Solution
A network device architecture that includes a forwarding plane, a control plane, and an AI circuit, where the forwarding plane pre-processes network packets to reduce the computational burden on the control plane and AI circuit, allowing for direct data transfer between the forwarding plane and AI circuit, thereby enhancing processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI algorithms are introduced into network devices for processing network packets, then network quality detection and analysis capabilities are improved, but processing efficiency deteriorates due to high computational demands overwhelming the forwarding plane
Solution Approach 1:
The system segments the data processing task into three distinct planes: the forwarding plane performs basic packet parsing and statistical data collection, the control plane handles data aggregation and format conversion, and the AI circuit performs complex AI algorithm execution. This segmentation allows each component to operate within its computational capabilities while collectively achieving efficient AI processing in network devices.
Solution Approach 2:
The control plane acts as an intermediary between the forwarding plane and the AI circuit. It receives statistical data from the forwarding plane, performs necessary data aggregation and format conversion, and then transfers the processed data to the AI circuit for analysis. This intermediary role prevents the AI circuit from being directly overwhelmed by raw packet data while maintaining efficient data flow.
2Productivity
If the forwarding plane performs comprehensive pre-processing of network packets, then the computational burden on the control plane and AI circuit is reduced, but the forwarding plane's calculation capability is overwhelmed
Solution Approach 1:
The pre-processing function is segmented across multiple planes: the forwarding plane performs basic packet parsing and extracts fundamental statistical data (packet count, byte count, flow duration), while the control plane handles more complex data aggregation and format conversion. This segmentation ensures that the forwarding plane does not undertake computationally intensive tasks that would overwhelm it.
Solution Approach 2:
The forwarding plane performs partial pre-processing by extracting only the essential statistical data needed for AI analysis, rather than attempting comprehensive pre-processing. This partial action approach allows the forwarding plane to reduce computational burden while still providing sufficient data to the control plane and AI circuit for effective analysis.
3Measurement precision
If more statistical data is collected and pre-processed before AI analysis, then the quality of AI-driven network analysis is improved, but the data processing overhead increases
Solution Approach 1:
The data collection and processing pipeline is segmented into distinct stages: the forwarding plane collects basic statistical data in real-time, the control plane aggregates and validates this data, and the AI circuit performs analysis. This segmentation allows each stage to optimize for its specific function, ensuring high analysis accuracy while minimizing overall processing time through specialized processing at each level.
Solution Approach 2:
The forwarding plane performs preliminary data collection and basic statistical computation continuously as packets arrive, so that when the control plane needs to transfer data to the AI circuit, the data is already prepared and aggregated. This preliminary action reduces the processing time required at later stages while maintaining comprehensive and accurate statistical data for AI analysis.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A network device and a method for processing data about network packets, which relate to the field of artificial intelligence, are disclosed. The network device includes a forwarding plane and an artificial intelligence AI circuit. The forwarding plane is coupled to the AI circuit. The AI circuit is configured to process data about network packets from the forwarding plane using a first AI algorithm. The forwarding plane may be directly coupled to the AI circuit, and the forwarding plane may further pre-process the network packets to obtain the data about the network packets.