Packet Classification via Cyclic Rule Set Splitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing packet classification methods result in oversized decision trees due to rule overlapping, leading to increased complexity and slower processing times, as they struggle to efficiently split and manage large rule sets in network devices.
Innovation Solution
The method involves cyclically splitting an initial rule set into smaller subsets based on global average overlap rates, building separate decision trees for each subset, and using a fast tree-building algorithm to reduce tree scale and generation time, with the FPGA performing lookups and prioritizing matching rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional packet classification methods are used to build decision trees from large rule sets, then complete rule coverage is achieved, but the decision tree scale becomes oversized and processing speed decreases
Solution Approach 1:
The patent divides a large rule set into multiple smaller rule sets by identifying and separating overlapping rules. Each smaller rule set is then independently processed to build compact decision trees. This segmentation prevents the exponential growth of decision tree size while maintaining complete rule coverage through systematic overlap identification and separation.
2Device complexity
If rule sets are not split and processed separately, then simpler processing logic is maintained, but decision tree scale increases and building time extends
Solution Approach 1:
The patent performs preliminary analysis of the rule set to identify overlapping rules before building decision trees. By pre-processing the rule set to separate overlapping rules and organize them into distinct groups, the subsequent decision tree construction becomes faster and more efficient, reducing overall building time while maintaining processing logic simplicity.
3Ease of manufacture
If overlapping rules are not prevented during rule set splitting, then rule set management remains simple, but decision tree scale becomes unmanageably large
Solution Approach 1:
The patent implements a feedback mechanism that monitors rule set overlaps during the splitting process. By continuously analyzing rule overlaps and adjusting the splitting strategy accordingly, the system prevents decision tree scale from becoming unmanageably large while maintaining relatively simple rule set management through automated overlap detection and separation.
4Productivity
If cyclic splitting based on global average overlap rates is implemented, then rule set optimization is improved, but processing complexity increases
Solution Approach 1:
The patent uses cyclic splitting based on global average overlap rates as a key parameter to optimize rule set organization. By iteratively adjusting the splitting parameters and reorganizing rule sets based on measured overlap rates, the system achieves improved packet classification efficiency while managing algorithmic complexity through systematic parameter-based control.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
In an example, a method for packet classification may include cyclically splitting an initial rule set to generate multiple rule subsets, performing tree building for each rule subset to obtain multiple decision trees. The cyclic splitting may select a target rule set having a highest global average overlap rate, from a rule set group which includes the initial rule set before the cyclic splitting and will include the multiple rule subsets after the cyclic splitting; split the target rule set according to a split point to obtain two rule sets; add the two rule sets into the rule set group to replace the target rule set; and continue to select a new target rule set from the rule set group, until the number of rule sets in the rule set group reaches a preset number.