Sorted-Table Binary Search Segmentation for One-Cycle Throughput
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional binary search algorithms on a single sorted table require a minimum of N clock cycles to find a matching data, leading to poor throughput and are not suitable for high-speed search applications like packet lookup, necessitating a solution that reduces search time without increasing memory usage.
Innovation Solution
The sorted table is divided into multiple smaller tables, allowing concurrent pipelined searches, where each iteration searches one table and outputs a match or a search address for the next table, reducing the time between consecutive searches to a single clock cycle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single sorted table is used for binary search, then memory usage is minimized, but search throughput is poor and requires N clock cycles
Solution Approach 1:
The sorted table is divided into multiple smaller sorted tables (segments), where each table contains a subset of the original data. This segmentation allows multiple searches to be performed concurrently on different tables, transforming a sequential N-clock-cycle process into a pipelined operation that completes in one clock cycle per search after initial setup.
2Productivity
If a single sorted table is used, then memory size is optimized, but the table must be locked to a single search reducing concurrency
Solution Approach 1:
The single sorted table is segmented into multiple independent sorted tables, each capable of handling searches concurrently. This eliminates the need for locking mechanisms while maintaining data integrity, as each table operates independently with its own search pipeline.
3Speed
If the sorted table is broken into smaller sections for pipelining, then search speed improves to one clock cycle, but memory organization complexity increases
Solution Approach 1:
The memory is organized into multiple sorted table segments, each stored in separate memory locations. This physical segmentation enables parallel access and pipelined searches, achieving one-clock-cycle search speed while the memory controller manages the distributed structure through address mapping.
4Productivity
If conventional binary search is used on a single table, then implementation is simple, but throughput is very poor for high-speed applications
Solution Approach 1:
The search operation is segmented across multiple tables, with each table handling a portion of the search space. This allows the system to maintain simple binary search logic within each table while achieving high throughput through concurrent operations and pipelining across the segmented structure.
Data Source
AI summary
Pipelining a binary search algorithm of a sorted table is performed by breaking a single sorted table into multiple smaller tables, such that a binary search can produce a result every clock cycle. A process includes storing a plurality of tables that collectively form a single sorted table that includes addresses and entries corresponding to each address where the entries are sorted, each of the plurality of tables include a subset of the addresses and corresponding entries; and performing a binary search where each iteration of the binary search searches one of the plurality of tables such that a complete search traverses through all of the plurality of tables and such that a plurality of searches are performable concurrently pipelined through each of the plurality of tables.