Dual-Media SSD Cache for Small-Granularity File Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current storage systems face inefficiencies in handling small-granularity file updates and metadata writes, leading to increased latency and write amplification due to the need for read-modify-write operations on NAND-based solid-state drives, which can result in reduced performance and endurance.
Innovation Solution
Implementing a dual-media approach with a small-granularity, byte-addressable non-volatile memory (such as PCM or 3D crosspoint memory) as a write-back cache within the SSD, allowing for in-place writes and reducing the need to write updates to NAND media, along with a larger-granularity media for metadata and user data, utilizing a logical to physical translation table to manage data placement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If NAND-based solid state drives are used for storage, then non-volatile storage capacity is achieved, but latency increases and write amplification occurs due to read-modify-write operations for small-granularity file updates
Solution Approach 1:
The storage system is segmented into two distinct media types: a first non-volatile media with small write granularity for metadata and active file updates, and a second non-volatile media with large write granularity for bulk data storage. This segmentation allows small-granularity writes to be directed to the first media, avoiding the read-modify-write overhead of traditional NAND drives, while the second media handles large sequential writes efficiently.
Solution Approach 2:
A hybrid file system layer acts as an intermediary between the host and the dual-media storage system. It intelligently routes write operations to the appropriate media based on granularity requirements, managing data placement and movement between media without requiring host intervention, thereby reducing latency for small updates while maintaining high productivity for bulk operations.
2Reliability
If read-modify-write operations are performed on NAND-based solid state drives, then data updates are completed, but write amplification increases and drive lifetime is reduced
Solution Approach 1:
By segmenting the storage system into two media types with different write granularities, the patent eliminates the need for read-modify-write operations on NAND-based media. Small-granularity updates are directed to the first non-volatile media that supports in-place writes, preventing unnecessary read-modify-write cycles that cause write amplification and extend drive lifetime.
Solution Approach 2:
The first non-volatile media provides self-service capabilities for small-granularity writes by supporting in-place updates without requiring intervention from the host or complex file system operations. This self-service mechanism reduces the overall write amplification in the system by handling small updates efficiently within the media itself.
3Loss of time
If a dual-media approach with small-granularity non-volatile memory is implemented, then latency is reduced and write amplification decreases, but device complexity increases
Solution Approach 1:
The hybrid file system provides multi-functionality by simultaneously managing two different non-volatile media types with different write granularities. It handles both small-granularity metadata updates and large-granularity data writes through a unified interface, reducing the need for separate management systems and minimizing the increase in device complexity while achieving low latency.
Solution Approach 2:
The storage controller automatically manages data placement and movement between the two media types without requiring complex host-level intervention. This self-service mechanism simplifies the overall system architecture by embedding the complexity within the storage device itself, reducing the burden on the host system while achieving low latency for small-granularity operations.
Data Source
AI summary
An embodiment of a mass storage apparatus may include first non-volatile media, second non-volatile media which provides a relatively larger write granularity as compared to the first non-volatile media, and logic communicatively coupled to the first and second non-volatile media to direct an access request to one of the first non-volatile media and the second non-volatile media based on an indication from an operating system. An embodiment of a host computing apparatus may include a processor, memory communicatively coupled to the processor, and logic communicatively coupled to the processor to provide an indication for a file system-related access request to a mass storage device based on a granularity size for the file system-related access request. Other embodiments are disclosed and claimed.