Core Concepts
Evaluating the performance of E2LSH on modern storage devices and determining the storage requirements for optimal speed.
Abstract
The content discusses the implementation and evaluation of E2LSH on modern storage devices, focusing on its computational and I/O costs. It explores the impact of different storage interfaces and devices on query times, emphasizing the need for efficient external memory execution. The analysis includes experimental setups, datasets used, and comparisons with small-index LSH methods like SRS and QALSH.
Introduction to Locality Sensitive Hashing (LSH) for approximate nearest neighbor search.
Comparison of E2LSH with small-index LSH variants in terms of query time efficiency.
Experimental setup details including datasets, computational costs, and I/O requirements.
Analysis of required IOPS for achieving speeds comparable to in-memory implementations.
Examination of CPU overheads and storage interface impacts on query performance.
Stats
Since then, several LSH variants having much smaller index sizes have been proposed.
Our analysis indicates that external memory needs to provide a random read performance of a few hundred kIOPS in order to compete with the state-of-the-art small-index methods SRS [35] and QALSH [18] executed in-memory.
Quotes
"Our analysis indicates that external memory needs to provide a random read performance of a few hundred kIOPS." - Content