Multi-Level Hash Tables for FPGA Throughput
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Field Programmable Gate Arrays (FPGAs) have limited throughput in hash table operations due to small memory blocks and clock frequency limitations, which restricts the number of data entries that can be processed per clock cycle.
Innovation Solution
The implementation of multi-level hash tables (MHTs) and nested multi-level hash tables (NMHTs) in FPGAs, where multiple hash functions are used to distribute keys across multiple buffers and storage destinations, allowing for parallel processing and increased throughput by exploiting the parallel computation capabilities of FPGAs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If hash tables are built in FPGA memory blocks, then storage capacity is improved, but throughput is limited by clock frequency
Solution Approach 1:
The hash table is divided into multiple levels (top-level, middle-level, bottom-level hash tables) with multiple buckets at each level. This segmentation allows parallel processing of multiple keys simultaneously across different buckets, overcoming the throughput limitation imposed by single-clock-cycle operations in FPGA memory blocks.
Solution Approach 2:
The patent introduces a multi-level hierarchical structure adding depth dimension to the traditional single-level hash table. Keys are distributed across multiple levels and buckets, enabling parallel computation paths and increasing throughput by processing multiple keys in parallel rather than sequentially within a single clock cycle.
2Productivity
If multiple hash functions are used to distribute keys across multiple buffers, then throughput is improved, but device complexity increases
Solution Approach 1:
The system is segmented into distinct functional components: top-level hash tables for initial key distribution, middle-level hash tables for further distribution, bottom-level hash tables for final storage, and input buffers for key reception. Each component uses simplified hash functions appropriate to its level, reducing individual complexity while achieving high overall throughput through coordinated parallel operation.
Data Source
AI summary
Described herein are hash table structures and system and methods for creating and storing data in the hash table structures. Hash tables and input buffers associated with same are configured in several arrangements, including multi-level hash tables and nested multi-level hash tables. The hash table structures described herein are well suited for being configured in the memory blocks of field-programmable gate arrays.