Parallel Hash Flow Classification for High-Speed Network Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network switching technologies, particularly in core routers, face challenges in achieving high-speed Access Control List (ACL) lookup rates due to power consumption and memory overhead issues with existing TCAM-based methods, especially with the increased complexity and size of IPv6 rule sets, which require multiple TCAMs and result in reduced performance and increased power consumption.
Innovation Solution
A parallel hash lookup-based flow classification method and device that classifies rules into subsets based on selected characteristics, using hash storage units in a storage array, allowing for serial pipeline and parallel lookup, thereby reducing memory requirements and power consumption while maintaining high-performance table lookup.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If multiple TCAMs are scheduled for lookup to meet 400Gbps lookup rate requirement, then lookup speed is improved, but power consumption and heat dissipation become serious problems
Solution Approach 1:
The patent segments the rule set into multiple rule subsets based on matching characteristics (exact match, prefix match, mask match, range match). Each subset is processed by dedicated hardware modules (EM table, PM table, MM table, RM table), allowing parallel lookup operations that achieve high speed without requiring multiple TCAMs, thus reducing power consumption.
Solution Approach 2:
The patent replaces the TCAM-based mechanical lookup system with a hybrid architecture combining software-based rule classification (using tree-like structures and heuristic algorithms) with hardware-based parallel lookup tables. This substitution eliminates the need for multiple power-consuming TCAMs while maintaining high lookup rates through coordinated software-hardware processing.
2Adaptability or versatility
If TCAM-based lookup is used to support high-capacity rule sets, then lookup functionality is improved, but memory requirements and cost increase significantly
Solution Approach 1:
The patent segments the large rule set into four distinct rule subsets based on matching characteristics. Each subset is stored in a specialized hardware table (EM, PM, MM, or RM table) with appropriate data structure, rather than storing all rules in a single large TCAM. This segmentation reduces the memory capacity required in each hardware component while maintaining support for high-capacity rule sets overall.
Solution Approach 2:
The patent changes the parameters of rule representation and storage by classifying rules into different subsets with specific characteristics. Each subset uses optimized data structures and lookup mechanisms tailored to its matching type, allowing efficient storage and retrieval without requiring excessive memory resources that would be needed for a unified TCAM approach.
3Speed
If tree-like structures with rule copying are used for rapid lookup, then lookup speed is improved, but storage space increases geometrically with rule set size
Solution Approach 1:
The patent segments rules into four subsets based on matching characteristics and processes each subset through dedicated hardware tables. This segmentation eliminates the need for extensive rule copying required by tree-like structures, as each rule appears in only its appropriate subset rather than being replicated across multiple tree nodes, thus reducing storage space while maintaining lookup speed.
Solution Approach 2:
The patent minimizes rule copying by using a classification approach where each rule is placed in a single appropriate subset based on its matching characteristics. This contrasts with tree-like structures that require copying rules to multiple nodes, thereby reducing storage overhead while maintaining rapid lookup capability through direct hardware access to the relevant subset.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A flow classification method and device and a storage medium are provided. The method includes: multiple rules in a rule set are classified to obtain one or more rule subsets according to one or more selected characteristics (101); hash storage is performed according to each classified rule subset (102); and during rule lookup, hash lookup in each parallel hash storage unit is performed according to hash key values of the rules (103).