SSD Virtual Stream Mapping via ML Distance Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Flash memory-based SSDs face limitations in managing multiple virtual streams efficiently due to restrictive hardware capabilities, leading to suboptimal performance and reduced lifetime.
Innovation Solution
A storage device and operation method that map virtual streams to physical streams based on distance information, using a machine learning model to extract representative values and calculate similarity, thereby assigning the most suitable physical stream for each virtual stream to improve operation efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional stream management methods are used in flash memory-based SSDs, then hardware complexity is reduced, but stream management effectiveness and performance deteriorate due to restrictive hardware capabilities
Solution Approach 1:
The patent introduces a machine learning model as an intermediary component between the host and flash memory devices. This model extracts representative values from stream features and calculates distance information to map virtual streams to physical streams, enabling effective stream management without requiring complex hardware modifications. The machine learning model acts as a software-based mediator that bridges the gap between limited hardware capabilities and advanced stream management requirements.
2Measurement precision
If virtual streams are mapped to physical streams without considering similarity, then mapping speed is improved, but data management accuracy and performance deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model with stream feature data before actual stream mapping operations. The model learns to extract representative values and calculate distance information in advance, enabling accurate virtual-to-physical stream mapping without time-consuming computations during runtime. This pre-processing approach ensures both high accuracy and efficient mapping performance.
Solution Approach 2:
The patent uses copying by creating representative value copies from stream features that capture essential characteristics without duplicating entire data sets. The machine learning model extracts compact representative values that preserve the similarity information needed for accurate stream mapping, reducing computational overhead while maintaining mapping precision.
3Quantity of substance
If multiple virtual streams are managed with limited physical streams, then storage capacity is improved, but stream assignment accuracy deteriorates due to hardware limitations
Solution Approach 1:
The patent applies parameter changes by transforming stream features into representative values through machine learning processing. The system changes the parameter representation from raw stream data to compressed feature vectors, and further to distance metrics that quantify similarity between virtual and physical streams. This parameter transformation enables accurate stream assignment even when the number of virtual streams exceeds physical stream capacity.
Data Source
AI summary
A storage device is configured to manage a plurality of nonvolatile memories with a plurality of physical streams. An operation method of the storage device includes receiving an input/output request from an external host device, determining a 0-th virtual stream identifier, extracting a 0-th representative value from a 0-th virtual stream feature, extracting a first and second representative values corresponding to first and second physical streams, calculating distance information including first and second similarities between the 0-th virtual stream and each of the first and second physical streams, based on the extracted representative values, assigning one of the plurality of physical streams to the 0-th virtual stream, based on the distance information, and performing an operation corresponding to the input/output request, at the assigned physical stream, and the extracting and the calculating are performed by using machine learning model.