ZNS Victim Zone Selection via Data Overwrite Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In ZNS devices, the existing victim zone selection method based on valid page count (VPC) is inefficient for partially written zones, as it can lead to improper selection of zones for garbage collection, resulting in suboptimal memory reclaim and increased write amplification factor.
Innovation Solution
The controller selects victim zones for garbage collection based on the amount of data overwrites rather than VPC, by tracking the number of overwrites for each zone and grouping zones based on overwrite percentages, ensuring that zones with the highest overwrite amounts are prioritized for compaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If victim zones are selected based on valid page count (VPC), then zones with low VPC are prioritized for garbage collection, but this leads to improper selection for partially written zones and suboptimal memory reclaim
Solution Approach 1:
The patent changes the selection parameter from valid page count (VPC) to data overwrite amount. Instead of using VPC to identify victim zones, the system tracks and uses the number of overwrites for each zone, fundamentally altering the criterion for garbage collection priority to better handle partially written zones.
Solution Approach 2:
The patent replaces the traditional VPC-based mechanical selection process with an overwrite-tracking mechanism. By substituting the selection criterion from page count to overwrite frequency, the system achieves more accurate victim zone identification that accounts for partial write scenarios.
2Device complexity
If zones are selected for garbage collection without considering overwrite amounts, then processing is simpler, but write amplification factor increases due to suboptimal memory reclaim
Solution Approach 1:
The patent implements preliminary tracking of data overwrite amounts for each zone before garbage collection is needed. By maintaining overwrite counters and grouping zones based on their overwrite percentages in advance, the system prepares optimization data that reduces write amplification during actual garbage collection operations.
Solution Approach 2:
The system uses feedback from tracked overwrite amounts to dynamically adjust garbage collection priorities. Zones with higher overwrite percentages are identified and prioritized for compaction, creating a feedback loop that continuously optimizes memory reclaim efficiency and reduces write amplification factor.
3Quantity of substance
If victim zone selection is optimized for partially written zones, then memory reclaim is maximized, but the selection process becomes more complex
Solution Approach 1:
The patent segments zones into different groups based on their overwrite percentages. By dividing zones into categories (e.g., high overwrite, medium overwrite, low overwrite groups), the system can apply targeted garbage collection strategies to each segment, maximizing memory reclaim while managing complexity through structured classification.
Solution Approach 2:
The patent applies different selection criteria and processing approaches to different zone groups based on their local characteristics (overwrite amounts). High overwrite zones receive priority compaction while low overwrite zones are handled differently, allowing optimization tailored to each zone's specific state rather than applying a uniform approach.
Data Source
AI summary
Aspects of a storage device are provided including improved victim zone selection for zone-based GC in ZNS. The storage device includes a NVM having first, second, and third blocks, and a controller. The controller creates a first superblock including the first blocks, a second superblock including the second blocks, and a third superblock including the third blocks. The controller determines whether a quantity of data overwrites associated with a first zone in the first superblock is larger than a quantity of data overwrites associated with a second zone in the first superblock. From the determination, the controller performs GC for the first zone prior to GC for the second zone, in which data overwrites associated with the first zone in the first superblock and sequential data associated with the first zone in the third superblock are relocated to the second superblock. Thus, storage device performance is improved and WAF reduced.