Host Index Management for Key-Value Storage I/O Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage systems face inefficiencies in input/output throughput between hosts and key-value storage devices due to the need for block-sized updates and serial node rewriting, which reduces performance and increases resource usage.
Innovation Solution
A host system that communicates with a key-value storage device using a tree structure with index management and device drivers to generate logs for updates, allowing for efficient storage and retrieval of data without requiring block-sized updates, thereby reducing unnecessary input/output operations and resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If block-sized updates and serial node rewriting are used to manage key-value storage, then data integrity is maintained, but input/output throughput decreases and resource usage increases
Solution Approach 1:
The patent segments the index structure into multiple independent nodes organized in a tree hierarchy. Each node can be updated independently without requiring serial rewriting of the entire index, thereby improving I/O throughput while maintaining data integrity through the tree structure's inherent consistency guarantees
Solution Approach 2:
The patent implements preliminary actions by maintaining in-memory copies of index nodes and preparing update operations before committing them to storage. This allows batch processing of multiple updates and reduces the frequency of I/O operations, improving throughput while ensuring data integrity through controlled commit mechanisms
2Stability of the object's composition
If block-sized updates are performed on storage devices, then data consistency is ensured, but device lifetime is reduced due to increased write operations
Solution Approach 1:
The patent merges multiple small update operations into larger batch operations that are committed together. By combining multiple updates into single I/O operations, the total number of write operations to the storage device is reduced, extending device lifetime while maintaining data consistency through atomic batch commits
Solution Approach 2:
The patent implements periodic action by deferring commit operations until necessary and batching updates over time. Instead of immediately committing every update, the system accumulates changes and commits them periodically, reducing the frequency of write operations to the storage device and thereby extending its lifetime while maintaining consistency
3Manufacturing precision
If serial node rewriting is used to update the index tree, then correctness of the index structure is maintained, but processing speed decreases
Solution Approach 1:
The patent segments the index update process into independent node operations. Each node in the tree can be updated independently and in parallel where possible, eliminating the serial dependency chain. This segmentation maintains structural correctness through the tree's hierarchical organization while dramatically improving processing speed through parallel execution
Solution Approach 2:
The patent implements dynamics by allowing the index structure to be modified in-place with optimistic concurrency control. Nodes can be updated dynamically without requiring complete serial rewriting, and conflicts are resolved through versioning and validation mechanisms that preserve correctness while enabling faster parallel updates
Data Source
AI summary
A host, a storage system, and an operating method of the host are provided. The host includes a host memory configured to store a tree structure including a leaf node and an index node, an index management module configured to manage an index based on the tree structure and generate a first log corresponding to the leaf node based on a first update request corresponding to a first key-value entry included in the leaf node, and a device driver configured to generate a first write command corresponding to the first log and transmit the generated first write command to a key-value storage device, so as to store the first log in the key-value storage device. The index management module is configured to generate a first new key-value entry, the first new-key value entry including a first value updated based on the first update request, as the first log.