Address Mapping With Mixed Translation Units to Reduce Write Amplification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory sub-systems face challenges in managing large non-volatile memory devices due to the rapid increase in the size of volatile memory needed for logical-to-physical mapping information, leading to increased write amplification and reduced reliability of non-volatile memory devices, especially in types like QLC memory.
Innovation Solution
Implementing an address mapping table that uses mixed translation unit sizes based on the number of input/output chunks, selecting smaller units for smaller requests and larger units for larger requests to reduce the number of entries in the logical-to-physical mapping table, thereby reducing unnecessary write amplification and volatile memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a single large translation unit size is used for all memory requests, then the number of mapping table entries is reduced, but write amplification increases for small requests
Solution Approach 1:
The translation unit size is segmented into multiple levels (e.g., 4KB, 16KB, 64KB) based on the size of the memory request. Small requests use smaller translation units while large requests use larger translation units, allowing the system to optimize between mapping table size and write amplification for different request patterns.
Solution Approach 2:
The translation unit size is made dynamic rather than fixed, allowing it to adapt based on the request characteristics. The system selects appropriate translation unit sizes from multiple available sizes according to the number of I/O chunks in each request, enabling flexible optimization of both mapping table entries and write amplification.
2Quantity of substance
If a single large translation unit size is used, then volatile memory requirements are reduced, but non-volatile memory reliability decreases due to unnecessary write operations
Solution Approach 1:
Multiple translation unit sizes are segmented and assigned based on request characteristics, preventing unnecessary writes to non-volatile memory while keeping volatile memory requirements manageable through efficient mapping table organization.
Solution Approach 2:
The translation unit size parameter is changed dynamically based on the number of I/O chunks in each request. This parameter adaptation allows the system to minimize write operations to non-volatile memory by selecting appropriate translation unit sizes, thereby improving reliability while maintaining acceptable volatile memory usage.
3Loss of energy
If smaller translation unit sizes are used for all requests, then write amplification is reduced, but the number of mapping table entries and volatile memory requirements increase
Solution Approach 1:
The translation unit size is made dynamic, selecting from multiple sizes (4KB, 16KB, 64KB) based on the request size. This dynamic selection reduces mapping table entries for large requests while maintaining small translation units for small requests, optimizing both write amplification and volatile memory usage.
Solution Approach 2:
The translation unit size parameter is adjusted based on the number of I/O chunks in each request. This parameter change strategy allows the system to use smaller translation units when needed to reduce write amplification while using larger translation units when appropriate to minimize mapping table size and volatile memory requirements.
Data Source
AI summary
A processing device in a memory sub-system receives a plurality of requests to write data to a non-volatile memory device, the plurality of requests comprising respective numbers of input/output (I/O) chunks of a fixed size, and performs a plurality of write operations to write the data to the non-volatile memory device using respective translation units, wherein the respective translation units comprise two or more different translation unit sizes selected based on the respective numbers of I/O chunks of the plurality of requests.