Network Packet Classification Using Multi-Index Partial Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network devices are unable to effectively classify packets based on partial matches with traffic flow criteria, missing opportunities to exploit known properties of packets even in the absence of a complete match with a policy.
Innovation Solution
A multivariant classification method and apparatus that generates multiple strings from packet data, mapping them to a common set of indices to select actions for processing, routing, or switching, allowing for classification and processing of packets even with partial matches by using a string generator and index allocator within a network switching device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single search term is used for packet classification, then the search process is simple, but the ability to exploit partial matches and classify new flows is lost
Solution Approach 1:
The patent divides the single search term into multiple separate search terms, each corresponding to a specific criterion (e.g., source address, destination address, protocol type). Each criterion is independently evaluated and assigned a weight, allowing the system to perform partial matches on individual criteria rather than requiring a complete match on a single concatenated term.
Solution Approach 2:
The patent transitions from a one-dimensional single-term search to a multi-dimensional search space where each criterion represents a dimension. The search process evaluates multiple dimensions simultaneously, combining results through weighted scoring to determine the best matching policy, thereby expanding the classification capability beyond the limitations of single-term matching.
2Measurement precision
If complete match criteria are enforced for policy application, then classification accuracy is high, but new and unknown traffic patterns cannot be classified
Solution Approach 1:
The patent applies partial matching by evaluating each criterion independently and assigning weights to indicate the importance of each match. Instead of requiring complete matches on all criteria, the system accumulates scores from partial matches and applies policies based on threshold thresholds, enabling classification of new traffic patterns that partially resemble known flows.
Solution Approach 2:
The patent introduces weight parameters for each criterion that can be adjusted to balance between classification accuracy and adaptability. By modifying these weight parameters, the system can adapt to different traffic patterns and policies, allowing flexible classification of new flows while maintaining accuracy for known flows through appropriate weight assignments.
Data Source
AI summary
A multivariant, common basis classification method and apparatus for classifying protocol data units in a network switching device is disclosed. The method of classifying a protocol data unit (PDU) in the preferred embodiment includes the steps of generating a first string and a second string with which to characterize a PDU; mapping first and second strings into a first index and second index, respectively, where the first and second indices are selected from a plurality of indices; and selecting an instruction to apply to the PDU by matching the first and second indices. The plurality of strings are commonly generated from the source and destination address in the packet as well as other indicia with which to identify a traffic flow. The instructions specify how to classify, route, switch, or otherwise process the PDU.