Asymmetric Memory Small Block Write Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing non-volatile memory systems, particularly flash memory, face inefficiencies in writing small and ultra-small blocks of data due to mismatched write unit sizes, leading to increased latency and write amplification, which results in wasted space and reduced performance.
Innovation Solution
Implementing a log-based storage structure where small blocks of data are initially written to a temporary storage and then reconciled into larger units for efficient writing, using a dynamic configuration of asymmetric memory to manage temporary and reconciled storage based on workload percentages and redundancy structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If small blocks of data are written directly to non-volatile memory using traditional methods, then the write operation can be performed, but write amplification increases and storage efficiency decreases due to mismatched write unit sizes
Solution Approach 1:
The patent segments the write operation into two distinct phases: (1) accumulating small data blocks in a log structure in memory, and (2) reconciling them into larger units for actual memory writing. This segmentation allows small writes to be buffered and combined, reducing write amplification while maintaining high write throughput for small blocks.
Solution Approach 2:
The patent introduces a log structure as an intermediary between the host and the non-volatile memory. This log acts as a buffer that temporarily stores small data blocks before they are reconciled into larger units suitable for efficient memory writing, thereby decoupling the small write operations from the larger physical write units.
2Loss of substance
If small blocks of data are written using log-based structure, then write amplification is reduced, but device complexity increases due to additional management overhead
Solution Approach 1:
The patent implements dynamic allocation and management of the log structure, where the log size and reconciliation thresholds are adjusted based on workload characteristics. The system dynamically determines when to switch between direct writing and log-based writing modes, optimizing performance while managing complexity through adaptive control rather than static configurations.
3Productivity
If asymmetric memory is used with different latency for read and write operations, then storage performance can be optimized, but managing small block writes becomes more complex
Solution Approach 1:
The patent performs preliminary actions by pre-allocating log space and pre-processing small data blocks before they need to be written to the asymmetric memory. The log structure is prepared in advance to accommodate small writes, and reconciliation operations are scheduled ahead of time, reducing the impact of asymmetric latency on overall performance.
Data Source
AI summary
A first portion of an asymmetric memory is configured as temporary storage for application data units with sizes corresponding to a small memory block that is smaller than the size of a logical write unit associated with the asymmetric memory. A portion of the remaining asymmetric memory is configured as a reconciled storage for application data units with varying sizes. A first application data unit is received for writing to the asymmetric memory. Based on computing the size of the first application data unit as corresponding to the small memory block, the first application data unit is written to the temporary storage. Upon determining that a threshold is reached, a memory write operation is performed for writing the application data units from the temporary storage to the reconciled storage. The application data units written to the reconciled storage are removed from the temporary storage.