LSM-Tree Storage Segmentation for KV SSD Space Amplification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The log-structured merge-tree (LSM-Tree) database storage architecture experiences severe space amplification due to append-only writing, leading to shortened SSD life and increased storage costs, while leveraging its fast writing performance and SSTable file storage.

Innovation Solution

A data access method that generates an index for key-value pairs, storing key-index pairs in a block storage device using the LSM-Tree structure and index-value pairs on a key-value solid state drive (KV SSD), allowing for sequential writing and reducing space amplification by avoiding in-place updates.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If append-only writing mechanism is used in LSM-Tree for block storage device, then writing performance is improved, but space amplification increases severely

Engineering Contradiction:
Improvewriting performanceVSAvoidspace amplification
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent segments the storage system into two distinct components: a block storage device (HDD or block SSD) for storing key-index pairs using LSM-Tree structure, and a key-value SSD for storing index-value pairs. This segmentation allows each storage device to handle specific data types optimally, enabling the LSM-Tree to maintain its sequential writing advantage while the key-value SSD handles updates efficiently, thereby reducing space amplification.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts the value storage function from the traditional LSM-Tree structure and places it in a separate key-value SSD. Only key-index pairs are stored in the block storage device using append-only writing, while index-value pairs are stored separately. This extraction eliminates the need for in-place updates in the block storage device, maintaining writing performance while reducing space amplification caused by invalid data retention.

Inventive Principle:
Principle #2Taking out (Extraction)

2Productivity

If append-only writing is used instead of in-place update, then writing performance is improved, but service life of SSD is shortened

Engineering Contradiction:
Improvewriting performanceVSAvoidservice life of SSD
Core Design Contradiction:
ProductivityVSDuration of action of stationary object

Solution Approach 1:

The patent divides the storage system into block storage device for key-index pairs and key-value SSD for index-value pairs. The block storage device uses append-only writing which is efficient but has limited write endurance, while the key-value SSD handles the update-intensive value storage. This segmentation protects the block storage device from excessive write operations, extending its service life while maintaining high writing performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts the update operations from the block storage device by storing value data in a separate key-value SSD. Since key-value SSDs are optimized for handling updates and have better write endurance, this extraction allows the block storage device to use efficient append-only writing without suffering from shortened service life, as the high-write operations are offloaded to the key-value SSD.

Inventive Principle:
Principle #2Taking out (Extraction)

3Productivity

If all writing operations are sequential append operations, then writing performance is improved, but data access complexity increases

Engineering Contradiction:
Improvewriting performanceVSAvoiddata access complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments data access into two distinct phases: first, searching for the key in the LSM-Tree structure in the block storage device to obtain the index, and second, using the index to retrieve the value from the key-value SSD. This segmentation of access operations simplifies the writing process (maintaining sequential append efficiency) while providing a clear, two-step access mechanism that balances complexity with performance.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS11537582B2Data access method, a data access control device, and a data access system
Publication Date: 2022.12.27 SAMSUNG ELECTRONICS CO LTD
  • US11537582B2 patent drawing
  • US11537582B2 patent drawing
  • US11537582B2 patent drawing

AI summary

Provided is a data access method including generating an index corresponding to a key in a first key-value pair in response to receiving the first key-value pair, writing the key-index pair corresponding to the key and the index into a block storage device, the key-index pair being stored in the block storage device in a log-structured merge-tree (LSM-Tree) structure, writing an index-value pair corresponding to the index and the value in the first key-value pair into a key-value solid state drive (KV SSD).