Data Plane Snoop-and-Learn Intelligence for Flow Table Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity in synchronizing control plane and data plane operations in network infrastructure devices leads to errors and inefficiencies, particularly in managing flow and L2 header information, due to the need for multiple command/response mechanisms and modifications across various protocol stacks.
Innovation Solution
The approach involves marking incoming packets, allowing the data plane to snoop and manage flow and L2 table updates without explicit command/response mechanisms, reducing the need for changes in control plane modules and simplifying the system by using packet descriptors to propagate flow and L2 header information, and employing dummy packets to maintain table synchronization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If control plane uses multiple command/response mechanisms to manage data plane flows and L2 headers, then synchronization accuracy is improved, but device complexity increases
Solution Approach 1:
The patent extracts the flow management functionality from the control plane and implements it directly in the data plane. The data plane autonomously creates, updates, and deletes flows using packet descriptors contained within the packets themselves, eliminating the need for separate command/response mechanisms between control and data planes.
Solution Approach 2:
The data plane is empowered to self-manage flow tables and L2 header information without requiring control plane intervention. When packets arrive, the data plane automatically extracts flow information from packet descriptors, updates its own flow tables, and manages L2 headers independently, making the system self-servicing.
2Measurement precision
If control plane periodically queries data plane for flow activity, then flow deletion accuracy is improved, but loss of time increases
Solution Approach 1:
Instead of periodic querying, the patent implements continuous flow activity tracking. The data plane continuously monitors packet arrivals and automatically updates flow timestamps as packets are processed, eliminating the need for periodic control plane queries and enabling immediate flow deletion when appropriate.
Solution Approach 2:
The patent implements feedback mechanisms where the data plane automatically reports flow status and activity information to the control plane only when necessary, rather than requiring continuous polling. This feedback-based approach maintains synchronization while reducing the time loss associated with periodic queries.
3Productivity
If data plane processes packets autonomously without control plane intervention, then productivity is improved, but reliability decreases
Solution Approach 1:
The patent embeds all necessary flow management information, including packet descriptors and L2 header data, directly within the packets before they reach the data plane. This preliminary preparation allows the data plane to autonomously process packets with complete information, eliminating the need for control plane intervention while maintaining synchronization reliability through pre-packaged accuracy.
Data Source
AI summary
An approach is provided where incoming packets are received at a data plane and header fields are extracted from the incoming packet. Flows from a flow data store are matched with the extracted header fields from the incoming packet. Packet descriptor data associated with the incoming packet is marked in the selected incoming packet forming a marked ingress packet with marking performed when the matching fails. The marked ingress packet is forwarded to a control plane that retrieves flow-related data related to the marked ingress packet and updates the marked packet descriptor data using the retrieved flow-related data, thereby forming an updated marked packet. The control plane passes the updated marked packet back to the data plane for further processing to update the flow data stored in the flow data store.