UFS SLC Buffer Resizing Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing caching techniques in universal flash storage (UFS) face issues with static and dynamic Turbo Write methods, leading to reduced SLC buffer lifetime, decreased writing performance, and suboptimal storage utilization due to fixed or adjusted SLC buffer sizes, resulting in deteriorated user experience and performance.
Innovation Solution
A caching method and apparatus utilizing a neural network, such as Mobilenet_V3, to dynamically adjust the size of the SLC buffer based on current storage and workload parameters, incorporating weight factors to optimize SLC buffer size according to real-time storage conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the size of the SLC buffer is increased to improve writing performance, then the writing speed improves, but the storage utilization rate decreases and the device cannot achieve optimal performance
Solution Approach 1:
The patent implements dynamic adjustment of the SLC buffer size based on real-time storage conditions. The controller monitors available storage capacity and workload characteristics, then adaptively modifies the SLC buffer size accordingly. This dynamic approach allows the system to optimize writing performance when storage capacity is abundant while maintaining high storage utilization when capacity is limited, resolving the contradiction between writing performance and storage utilization rate.
2Device complexity
If the size of the SLC buffer is fixed to simplify control, then the control complexity is reduced, but the lifetime of the SLC buffer decreases due to significant increase in erasing operations
Solution Approach 1:
The patent employs dynamic adjustment of SLC buffer size based on real-time monitoring of storage conditions and workload characteristics. The controller adaptively modifies the buffer size to minimize unnecessary erasing operations, thereby extending the SLC buffer lifetime while maintaining manageable control complexity through automated decision-making algorithms.
Solution Approach 2:
The system implements a feedback mechanism where the controller continuously monitors storage conditions, including available capacity and workload patterns, and uses this information to dynamically adjust the SLC buffer size. This closed-loop control optimizes the balance between buffer lifetime and writing performance by reducing redundant erasing operations while maintaining adaptive responsiveness to changing storage conditions.
3Quantity of substance
If the SLC buffer size is dynamically adjusted according to available storage capacity, then the storage utilization rate improves, but the writing performance declines sharply when the SLC buffer is fully written
Solution Approach 1:
The patent implements dynamic adjustment of SLC buffer size based on real-time storage conditions and workload characteristics. By monitoring available storage capacity and adjusting the buffer size adaptively, the system maintains optimal writing performance while maximizing storage utilization. The dynamic approach prevents the buffer from being consistently oversized or undersized, ensuring efficient use of storage resources without compromising writing speed.
Solution Approach 2:
The system changes the SLC buffer size parameter dynamically based on storage conditions and workload characteristics. This parameter adjustment allows the system to optimize the balance between storage utilization and writing performance by allocating appropriate buffer resources according to actual needs, preventing both excessive buffer sizes that waste storage capacity and insufficient buffer sizes that degrade writing performance.
Data Source
AI summary
There provides a caching method and apparatus of a universal flash storage (UFS). The caching method of the UFS includes: acquiring at least one current storage information of the UFS including a single level cell (SLC) buffer and a multi-level cell at predetermined interval, wherein the multi-level cell includes at least two levels of cells; acquiring input data of a neural network based on the at least one current storage information; acquiring adjustment information of the SLC buffer via the neural network based on the input data; and adjusting a size of the SLC buffer based on the adjustment information of the SLC buffer.