Multi-tier Memory Management with Dual-Copy Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current SSD storage devices face performance tradeoffs between write and read performance due to relocation policies that move data from fast tiers to slow tiers during idle time, leading to uncertainty in determining the optimal timing and magnitude of data relocation, which affects overall system efficiency.
Innovation Solution
A memory mapping scheme that preserves data in fast tiers by maintaining two copies of data in different memory tiers, allowing read access from the faster tier and enabling efficient relocation and erasure of data, reducing write amplification and improving read performance without complicating garbage collection algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is relocated from fast tier to slow tier during idle time, then write performance is improved, but read performance deteriorates
Solution Approach 1:
The system performs preliminary data copying from the fast tier to the slow tier during idle time, but does not immediately erase the original data. This preliminary action prepares the slow tier for future writes while preserving fast tier data for potential read acceleration, resolving the contradiction between write performance improvement and read performance maintenance.
Solution Approach 2:
The system dynamically determines whether to erase data from the fast tier based on workload characteristics. The erasure decision is not static but adapts to real-time conditions, allowing the system to optimize between write performance (by erasing) and read performance (by preserving) based on actual usage patterns.
2Productivity
If data is completely cleared from the highest tier during idle time, then write performance is maximized, but read performance for relocated data deteriorates
Solution Approach 1:
The system performs preliminary copying of data to the slow tier without immediately performing the erasure action from the fast tier. This delayed erasure allows the system to maintain fast tier data longer, enabling read acceleration while still preparing the slow tier for future writes, thus reducing read access time loss.
3Productivity
If aggressive relocation policy is used to ensure high write performance, then write speed is improved, but read performance for relocated data deteriorates
Solution Approach 1:
The system dynamically adjusts the aggressiveness of relocation based on workload characteristics. Instead of using a fixed aggressive policy, the system adapts its behavior to balance write speed improvement with read performance maintenance, making the relocation policy flexible rather than rigid.
Solution Approach 2:
The system uses workload characteristics as feedback to determine the appropriate relocation strategy. By monitoring actual usage patterns, the system can adjust its relocation aggressiveness to optimize both write speed and read performance, creating a closed-loop control system.
4Speed
If minimal relocation is performed during idle time, then read performance is maintained, but write performance acceleration is reduced
Solution Approach 1:
The system performs a partial relocation action by copying data to the slow tier without completing the full erasure from the fast tier. This partial action provides enough write performance improvement while maintaining sufficient fast tier data for read acceleration, avoiding the need for complete relocation.
5Speed
If data is stored in both tiers simultaneously, then read performance is improved through fast tier access, but device complexity increases due to dual mapping
Solution Approach 1:
The system extracts the erasure decision from the immediate relocation process. Instead of erasing fast tier data as part of the relocation operation, the erasure is separated and performed later based on independent workload analysis, simplifying the mapping management while maintaining read performance benefits.
Solution Approach 2:
The system performs preliminary copying to the slow tier without immediately performing erasure from the fast tier. This separation of actions reduces the complexity of coordinate mapping management while still achieving read performance improvement through fast tier data preservation.
Data Source
AI summary
The disclosure relates in some aspects to managing multi-tier memory, such as multi-tier NVM. Data that is originally written to a first tier (e.g., a fast tier) may be subsequently copied to a second tier (e.g., a slow tier). The data is temporarily left in the first tier until the space is needed for a subsequent write operation. Thus, for a period of time, a read operation is able to read the data from the first tier (e.g., the fast tier) instead of the second tier (e.g., the slow tier), thereby improving read performance. The disclosure relates in some aspects to a memory mapping scheme that enables a read operation to readily determine that data remains in the first tier and locate the data in that tier. Moreover, the scheme enables efficient reconfiguration of the mapping when the data in the first tier is erased.