Graph Partitioning for Scalable Longest Prefix Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Ternary Content Addressable Memories (TCAMs) face challenges with scalability and power consumption, especially when dealing with ternary keys containing wildcards, and algorithmic approaches struggle with chaotic wildcard distributions, limiting their performance and scalability in search applications.
Innovation Solution
The proposed solution involves partitioning ternary keys into sub-tables represented as sub-graphs, where each sub-table is searched in parallel by a fixed number of execution threads, optimizing the architecture for longest prefix matching (LPM) by locating specified bits in most significant positions and wildcards in least significant bits, and constructing graphs that require only single node processing for lookups, with the ability to handle wildcard bits efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If TCAM is used for ternary key search, then search performance is high and deterministic, but power consumption is extreme and scalability is limited
Solution Approach 1:
The patent divides the TCAM into multiple sub-TCAMs, each handling a specific subset of ternary keys. This segmentation allows the system to maintain high search performance for each subset while reducing the overall power consumption compared to a single large TCAM, and enables scalability by adding more sub-TCAMs as needed.
Solution Approach 2:
The patent changes the operating parameters by introducing a threshold parameter that determines when to use TCAM versus algorithmic search. By dynamically adjusting this threshold based on key characteristics (number of wildcards), the system optimizes the balance between power consumption and search performance.
2Use of energy by stationary object
If algorithmic approach is used for ternary key search, then power consumption is reduced, but performance deteriorates when wildcards are distributed chaotically
Solution Approach 1:
The patent uses the number of wildcards in a key as a parameter to determine the search strategy. When the number of wildcards is below a threshold, algorithmic search is used (low power). When the number of wildcards exceeds the threshold, TCAM search is used (high performance). This parameter-based decision optimizes both power consumption and performance.
Solution Approach 2:
The patent dynamically switches between algorithmic search and TCAM search based on the characteristics of the query key. This dynamic adaptation allows the system to optimize performance and power consumption in real-time, rather than using a fixed approach.
3Quantity of substance
If TCAM scales to larger capacity, then more keys can be stored, but scalability is limited to a few hundred thousand keys
Solution Approach 1:
The patent segments the large TCAM into multiple smaller sub-TCAMs, each manageable in size. This allows the system to scale by adding more sub-TCAMs rather than building a single enormous TCAM, improving scalability while maintaining the ability to store large numbers of keys.
Solution Approach 2:
The patent adds a dimensional aspect by introducing multiple levels of search (algorithmic search for some keys, TCAM search for others). This multi-dimensional approach to search allows the system to handle larger key spaces than a single TCAM level could accommodate.
4Productivity
If graph partitioning is applied to sub-tables, then computational overhead is reduced and parallel processing is enabled, but device complexity increases
Solution Approach 1:
The patent segments the search system into multiple independent sub-TCAMs that can be processed in parallel. Each sub-TCAM handles a specific partition of keys, allowing concurrent search operations that reduce overall computational overhead and improve productivity.
Solution Approach 2:
The patent performs preliminary partitioning of keys into subsets that are assigned to different sub-TCAMs. This preliminary organization enables efficient parallel processing during search operations, as each sub-TCAM can independently process its assigned subset without coordination overhead.
Data Source
AI summary
Methods, apparatus, and systems for efficient partitioning and construction of graphs for scalable high-performance search applications. In one aspect a graph-based method for performing a longest prefix match (LPM) is disclosed. A plurality of ternary keys and created or accessed, each representing an Internet Protocol (IP) mask and having a length w and a number of specific bits comprising a prefix length followed by one or more wildcards. The ternary keys are partitioned into subsets as a function of the prefix lengths of the ternary keys. For each subset, a graph is constructed, and the graph is stored in memory. The graphs are searched for a match for an IP address. A result associated with the graph associated with the subset of prefixed with the longest prefix length is returned. Associated apparatus and systems for implementing the methods are also disclosed. In some embodiments, a graph memory engine (GME) is used.