Zone-based mapping reduces DRAM overhead in SSDs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional memory sub-systems face high costs and performance losses due to random allocations of data, resulting in significant mapping overhead and write overhead, especially in solid-state drives, where gigabytes of DRAM are required for logical-to-physical address mapping, and periodic snapshotting and logging are necessary to manage data locality.
Innovation Solution
Implementing dynamic data placement and mapping that writes data groups sequentially into media blocks, increasing mapping granularity from kilobytes to hundreds of megabytes, allowing for reduced mapping overhead and parallel or serial writing across multiple IC dies, thereby simplifying the tracking of data locations and reducing collisions during concurrent access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If random allocations are used for data storage, then data can be stored flexibly, but mapping overhead increases significantly requiring gigabytes of DRAM
Solution Approach 1:
The patent segments data into data groups with data locality and assigns each group to a specific zone in the logical address space. This segmentation allows the system to manage mapping at a coarser granularity (zone level rather than individual block level), reducing the amount of mapping memory required while maintaining flexible data storage capabilities.
Solution Approach 2:
The patent changes the granularity parameter of address mapping from fine-grained (individual data blocks) to coarse-grained (data groups at zone level). This parameter change reduces mapping overhead from gigabytes to kilobytes of DRAM while preserving data storage flexibility through the zone-based organizational structure.
2Measurement precision
If fine-grained logical-to-physical address mapping is used, then data access precision is improved, but mapping overhead increases to gigabytes per terabyte
Solution Approach 1:
The patent introduces an intermediate segmentation level (zones) between logical and physical address spaces. Data groups are mapped to zones, and zones are mapped to physical media blocks. This multi-level segmentation maintains address mapping precision for data access while reducing overall mapping overhead by aggregating multiple data blocks into zone-level mappings.
Solution Approach 2:
The patent adds a hierarchical dimension to address mapping by introducing zones as an intermediate layer. Instead of direct one-to-one mapping, the system uses a three-level hierarchy (data groups → zones → physical blocks), which reduces the dimensionality of the mapping problem and decreases memory requirements while preserving access precision.
3Reliability
If periodic snapshotting and logging is performed, then data locality is managed, but write overhead increases and performance is reduced
Solution Approach 1:
The patent performs preliminary organization of data into groups with data locality before writing to media. By pre-establishing the zone-based structure and data group assignments, the system eliminates the need for periodic snapshotting and logging operations, thereby maintaining data locality management while avoiding the performance penalty of repeated write operations.
Solution Approach 2:
The patent extracts the data locality management function from the periodic snapshotting and logging process and integrates it into the primary write path through zone-based mapping. This extraction eliminates redundant operations, as data locality is managed inherently through the mapping structure rather than through separate periodic maintenance operations.
4Quantity of substance
If data is written sequentially into media blocks, then mapping overhead is reduced, but data access patterns must change
Solution Approach 1:
The zone-based mapping structure serves multiple functions simultaneously: it reduces mapping overhead, enables sequential writing patterns, and maintains efficient data access. The universal zone structure can accommodate both sequential and random access patterns while keeping mapping memory requirements low, making the system adaptable to different workloads without requiring separate optimization.
Data Source
AI summary
A system includes integrated circuit (IC) dies having memory cells and a processing device, which is to perform operations including generating a number of zone map entries for zones of a logical block address (LBA) space that are sequentially mapped to physical address space of the plurality of IC dies, wherein each zone map entry corresponds to a respective data group that has been sequentially written to one or more IC dies; and generating a die identifier and a block identifier for each data block of multiple data blocks of the respective data group, wherein each data block corresponds to a media block of the plurality of IC dies.