Application-Aware NVM Storage Segmentation for Write Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current non-volatile memory (NVM) storage systems do not account for application-specific aspects, leading to increased garbage collection write operations, reduced endurance, and higher latency in NVM devices.
Innovation Solution
Classify incoming data objects based on application metadata and garbage collection metadata, assigning them to specific storage sets with dedicated garbage collection mechanisms to optimize storage and minimize write amplification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional garbage collection mechanisms are used without application-aware classification, then the storage system can handle general write operations, but the number of GC write operations increases and device endurance decreases
Solution Approach 1:
The patent segments the storage space into multiple storage sets (first storage set, second storage set, etc.) based on application metadata characteristics such as seriality and temperature. Data objects are classified and directed to appropriate storage sets, allowing differentiated GC strategies. This segmentation reduces unnecessary GC writes by isolating frequently rewritten data from stable data, thereby improving device endurance while maintaining efficient GC operations.
Solution Approach 2:
The patent applies local quality by implementing application-specific GC policies for different storage sets. The first GC policy is applied to the first storage set (containing data with certain characteristics), while a second GC policy is applied to the second storage set (containing data with different characteristics). This localized approach optimizes GC operations for each data type, reducing overall GC write amplification and improving reliability.
2Reliability
If application-specific storage management is implemented, then GC write operations are reduced and endurance improves, but system complexity increases due to classification and multiple storage sets
Solution Approach 1:
The patent performs preliminary classification of data objects into different storage sets based on application metadata (seriality, temperature, etc.) before write operations occur. This advance classification enables the system to apply appropriate GC policies proactively, avoiding the need for complex real-time decisions during GC operations. The classification metadata is stored and reused, simplifying ongoing management while maintaining improved endurance.
3Loss of time
If data objects are segregated into multiple storage sets, then GC write operations are minimized and latency decreases, but the storage management process becomes more complex
Solution Approach 1:
The patent segments storage into multiple sets based on data characteristics (seriality, temperature), allowing GC operations to be performed independently on each segment. This segmentation enables parallel GC processing and reduces the latency impact on unrelated data. The classification framework, while adding some complexity, provides a systematic approach that simplifies the management of segmented storage compared to ad-hoc solutions.
Data Source
AI summary
A system and method of managing storage on non-volatile memory (NVM) storage media, by at least one processor, may include receiving, from at least one client computing device, one or more data write requests, associated with application metadata, to store one or more respective data objects on the NVM storage media; performing a first classification of the one or more data objects, based on the application metadata, so as to associate each data object to a group of data objects; storing the data objects of each group in a dedicated storage set of a logical address space; and transmitting, or copying the data objects of each storage set to be stored in a respective, dedicated range of the NVM storage media.