Packet Classification via Multi-Dimensional Splitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing packet classification methods face challenges in achieving efficient search speed and reducing memory usage, particularly in network functions like firewalls, due to high memory access times and rule replication issues.
Innovation Solution
The FastSplit method constructs a decision tree structure by splitting the search space along rule boundaries and conducting k-splits on multiple dimensions for each node, constraining memory usage within a threshold to enable fast memory access and reduce rule replication, using a geometric representation of the rule-set to create a search structure with nodes that store rules in leaf nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional packet classification methods are used, then comprehensive rule matching is achieved, but memory access time increases and search speed decreases
Solution Approach 1:
The patent segments the multi-dimensional search space into multiple orthogonal cuttings along different dimensions. Each cutting divides the rule set into smaller subsets that can be independently searched, reducing the time required to access and evaluate rules while maintaining complete coverage for accurate matching.
Solution Approach 2:
The patent transforms the traditional single-dimensional rule search into a multi-dimensional search space where rules are organized along multiple orthogonal axes. This dimensional transformation allows simultaneous progression through multiple rule attributes, significantly reducing search time while preserving matching accuracy through geometric intersection of dimensional constraints.
2Measurement precision
If complete rule sets are stored for accurate packet classification, then classification accuracy is maintained, but memory usage increases
Solution Approach 1:
The patent segments the complete rule set into multiple smaller rule subsets distributed across different dimensional cuttings. Each subset occupies less memory individually, yet the union of all subsets provides complete classification coverage, thereby reducing total memory usage while maintaining classification accuracy.
Solution Approach 2:
The patent evaluates multiple partial rule matches across different dimensional cuttings and combines them to determine the final classification. Rather than requiring one complete rule match, the system uses partial matches from multiple dimensions, reducing the memory footprint of individual rule entries while preserving overall classification accuracy.
Data Source
AI summary
A computer-implemented method classifies a packet received from a network. The method comprises receiving the packet having header information and a rule-set that indicates a plurality of actions to perform on the packet. A geometric representation of the rule-set is constructed having a plurality of dimensions and rule boundaries. At least one rule boundary and dimension of the geometric representation is split to form a search structure having a plurality of nodes. Each node in the plurality of nodes is constrained to a predetermined memory threshold. The search structure is searched using the header information to obtain a rule in the rule-set that indicates an action to take on the packet.