Method and device for dynamically optimizing hybrid solid state disk based on large language model

Abstract

The application relates to the technical field of hybrid solid state disks, and discloses a hybrid solid state disk dynamic optimization method and device based on a large language model, which systematically classifies the parameter space of a hybrid solid state disk, constructs a parameter sensitivity matrix based on the classified parameters, determines the influence difference parameters of the parameters on performance under different workloads, collects system state information of the hybrid solid state disk in real time, generates an optimized parameter configuration scheme based on a large language model according to the system state information, the influence difference parameters and historical configuration information. The application can realize real-time perception of work load changes and dynamic focusing on the most sensitive parameter combination under the current load, realize collaborative reasoning and adaptive adjustment among multiple parameters, significantly reduce the write amplification rate and operation delay, and solve the problem that the response efficiency, system stability and energy consumption performance of the hybrid SSD are poor under a dynamic work load in the prior art.

Description

Technical Field

[0001] This application relates to the field of hybrid solid-state drive technology, and for example to a method and apparatus for dynamic optimization of hybrid solid-state drives based on a large language model. Background Technology

[0002] Hybrid SSDs, as a versatile storage solution, achieve an optimal balance between capacity and performance by integrating various types of flash memory cells, such as Single-Level Cell (SLC), Multi-Level Cell (MLC), Triple-Level Cell (TLC), and Quad-Level Cell (QLC). These devices support multiple configurations to meet diverse needs for storage density, speed, and reliability. Their dynamic flexibility allows flash memory cells to switch between different modes, enabling hybrid SSDs to adapt to varying workloads and improve overall efficiency. This adaptability makes hybrid SSDs a key technology in personal computing, data center, and cloud storage environments.

[0003] However, the introduction of multiple flash memory modes brings new and more complex SSD management challenges, significantly impacting overall system performance. These challenges stem from the need to effectively handle the diverse flash memory types with unique performance, durability, and reliability characteristics. For example, data placement must be balanced between faster but lower-density SLC regions and slower but higher-capacity QLC regions, requiring sophisticated algorithms to achieve optimal resource utilization. Furthermore, the dynamic switching of flash cells between different modes adds complexity, as poor management can lead to increased write amplification, wear leveling imbalances, and excessive garbage collection overhead. These challenges are further exacerbated by changing workload demands, requiring continuous real-time adjustments to ensure system stability and maintain performance.

[0004] Therefore, among existing technical solutions, hybrid (Solid State Drive, SSD) generally performs poorly in terms of response efficiency, system stability, and energy consumption under dynamic workloads.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application. Summary of the Invention

[0006] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.

[0007] This disclosure provides a method, apparatus, device, and storage medium for dynamic optimization of hybrid solid-state drives based on a large language model, which solves the problem that hybrid SSDs generally have poor response efficiency, system stability, and energy consumption under dynamic workloads in existing technical solutions.

[0008] According to a first aspect of this disclosure, a dynamic optimization method for hybrid solid-state drives based on a large language model is provided, the method comprising: The parameter space of hybrid solid-state drives is systematically classified into hardware characteristic parameters, workload characteristic parameters, and policy control parameters. Based on the classified parameters, a parameter sensitivity matrix is ​​constructed to determine the differences in the impact of parameters on performance under different workloads. Real-time acquisition of system status information of hybrid solid-state drives, including I / O mode, flash memory health status, cache hit rate, space utilization, SLC mode ratio, garbage collection count, average operation latency, and write amplification rate; Based on the large language model, an optimized parameter configuration scheme is generated according to the system state information, influencing difference parameters, and historical configuration information.

[0009] Among some implementations of the first aspect, the above method also includes: The performance of the parameter configuration scheme is verified. If the performance improvement is lower than the preset threshold or the performance decline exceeds the preset ratio, the previous configuration is rolled back.

[0010] In some implementations of the first aspect, the aforementioned hardware characteristic parameters include garbage collection trigger threshold, wear leveling strategy, and space conversion granularity; The workload characteristic parameters include data heat classification period, read-write ratio, and I / O burst intensity; The policy control parameters include the clustering algorithm convergence threshold and the reinforcement learning training interval.

[0011] In some implementations of the first aspect, the aforementioned system status information is collected by a workload monitor, which uses a sliding window mechanism to calculate the variability of the workload. The standard deviation is calculated using the following formula: , in For window size, For the first One observation value, This represents the mean value within the window.

[0012] In some implementations of the first aspect, when generating optimized parameter configuration schemes based on large language models, a dynamic temperature parameter adjustment mechanism is introduced; When the workload variability is below the first threshold, the temperature parameter is set to a preset low temperature range, so that the large language model is biased to output a conservative strategy similar to the historical stable configuration. When the workload variability exceeds the second threshold, the temperature parameter is set to a preset high temperature range to increase the randomness and exploratory nature of the output configuration of the large language model in order to adapt to sudden load changes.

[0013] In some implementations of the first aspect, the above-mentioned optimized parameter configuration scheme includes: The large language model outputs a structured configuration list, which is then converted into a hardware-recognizable configuration file by the natural language parsing module, and subjected to format validation and numerical range constraint filtering.

[0014] Among some implementations of the first aspect, the above method also includes: The preset time after generating the configuration is used as the observation period, and the average performance during the corresponding period is calculated and compared with the previous period. If the performance drops by more than 5%, the previous configuration will be restored and configuration generation will be triggered again.

[0015] According to a second aspect of this disclosure, a hybrid solid-state drive dynamic optimization apparatus based on a large language model is provided for performing the method of the first aspect; the apparatus includes: The classification module is used to systematically classify the parameter space of hybrid solid-state drives into hardware characteristic parameters, workload characteristic parameters, and policy control parameters. Based on the classified parameters, a parameter sensitivity matrix is ​​constructed to determine the differences in the impact of parameters on performance under different workloads. The acquisition module is used to collect system status information of the hybrid solid-state drive in real time, including I / O mode, flash health status, cache hit rate, space utilization, SLC mode ratio, garbage collection count, average operation latency and write amplification. The optimization module is used to generate an optimized parameter configuration scheme based on the large language model, the system state information, the parameters affecting differences, and the historical configuration information.

[0016] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the program to implement the method of the first aspect.

[0017] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method of the first aspect.

[0018] The method, apparatus, device, and storage medium for dynamic optimization of hybrid solid-state drives based on large language models provided in this disclosure can achieve the following technical effects: This disclosure provides a method, apparatus, device, and storage medium for dynamic optimization of hybrid solid-state drives based on a large language model. It systematically classifies the parameter space of hybrid SSDs into three key parameters: hardware characteristics, workload features, and policy control. Utilizing the contextual reasoning and sequence modeling capabilities of the large language model, it constructs a closed-loop optimization framework that includes workload monitoring, thermal classification and space transformation, LLM automatic tuning, and performance verification. Through carefully designed prompting engineering, it transforms system states and historical configurations into natural language descriptions, guiding LLM to dynamically generate cross-parameter collaborative optimization configurations (such as linked adjustments to GC thresholds and slice sizes). Simultaneously, it introduces a dynamic temperature parameter adjustment mechanism to balance exploratory and conservative strategies, and configures error correction and performance rollback mechanisms to ensure system robustness. The technical advantages of this solution are as follows: Compared with traditional machine learning methods that only optimize a single subset and lack the ability to correlate global parameters, this invention can sense workload changes in real time and dynamically focus on the most sensitive parameter combination under the current load, realizing collaborative reasoning and adaptive adjustment among multiple parameters. This significantly reduces write amplification and operation latency, improves the response efficiency, system stability and energy consumption of hybrid SSDs under dynamic workloads, and its parameter hierarchical classification mechanism also provides a scalable architectural foundation for the integration of new flash memory types and management strategies in the future.

[0019] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description

[0020] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1 This is a flowchart illustrating a dynamic optimization method for hybrid solid-state drives based on a large language model, provided in an embodiment of this disclosure. Figure 2 This is a schematic diagram of a hybrid SSD internal architecture provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of the overall architecture of an LLM-SSD provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram of an LLM configuration tuning process provided in an embodiment of this disclosure; Figure 5 This is a schematic diagram of the structure of a hybrid solid-state drive dynamic optimization device based on a large language model provided in an embodiment of this disclosure; Figure 6 This is a schematic diagram of the structure of a hybrid solid-state drive dynamic optimization device based on a large language model, provided in an embodiment of this disclosure. Detailed Implementation

[0021] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0022] The terms "first," "second," etc., used in the embodiments of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0023] Unless otherwise stated, the term "multiple" means two or more.

[0024] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0025] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0026] The term "correspondence" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.

[0027] To address the aforementioned issues, this disclosure provides a method, apparatus, device, and storage medium for dynamic optimization of hybrid solid-state drives based on a large language model.

[0028] The following description, in conjunction with the accompanying drawings, illustrates the dynamic optimization method, apparatus, device, and storage medium for hybrid solid-state drives based on large language models provided in this disclosure.

[0029] Figure 1 This is a flowchart illustrating a dynamic optimization method for hybrid solid-state drives based on a large language model, provided in an embodiment of this disclosure.

[0030] like Figure 1 As shown, a dynamic optimization method for hybrid solid-state drives based on a large language model is proposed, which includes: S101 systematically classifies the parameter space of hybrid solid-state drives into hardware characteristic parameters, workload characteristic parameters, and policy control parameters, and constructs a parameter sensitivity matrix based on the classified parameters to determine the differences in the impact of parameters on performance under different workloads. S102 collects real-time system status information of the hybrid solid-state drive, including input / output (I / O) mode, flash health status, cache hit rate, space utilization, SLC mode ratio, garbage collection count, average operation latency, and write amplification. S103, based on a large language model, generates an optimized parameter configuration scheme according to system status information, influencing difference parameters, and historical configuration information.

[0031] In some embodiments, Figure 1 The methods may also include: The performance of the parameter configuration scheme is verified. If the performance improvement is less than the preset threshold or the performance decline exceeds the preset ratio, the previous configuration is rolled back.

[0032] In some embodiments, the above hardware characteristic parameters include garbage collection trigger threshold, wear leveling strategy, and space conversion granularity; Workload characteristics include data popularity classification period, read / write ratio, and I / O burst intensity; Policy control parameters include the clustering algorithm convergence threshold and the reinforcement learning training interval.

[0033] In some embodiments, the above system status information is collected by a workload monitor, which uses a sliding window mechanism to calculate the variability of the workload. The standard deviation is calculated using the following formula: , in For window size, For the first One observation value, This represents the mean value within the window.

[0034] In some embodiments, when generating an optimized parameter configuration scheme based on a large language model, a dynamic temperature parameter adjustment mechanism is introduced; When the workload variability is below the first threshold, the temperature parameter is set to a preset low temperature range, so that the large language model is biased to output a conservative strategy similar to the historical stable configuration. When the workload variability exceeds the second threshold, the temperature parameter is set to a preset high temperature range to increase the randomness and exploratory nature of the output configuration of the large language model in order to adapt to sudden load changes.

[0035] In some embodiments, the above-described optimized parameter configuration scheme includes: The large language model outputs a structured configuration list, which is then converted into a hardware-recognizable configuration file by the natural language parsing module, and subjected to format validation and numerical range constraint filtering.

[0036] In some embodiments, the above method may further include: The preset time after generating the configuration is used as the observation period, and the average performance during the corresponding period is calculated and compared with the previous period. If the performance drops by more than 5%, the previous configuration will be restored and configuration generation will be triggered again.

[0037] Figure 2 This is a schematic diagram of a hybrid SSD internal architecture provided in an embodiment of this disclosure. Figure 3 This is a schematic diagram of the overall architecture of a Large Language Model-Solid State Drive (LLM-SSD) disclosed herein. Figure 4 This is a schematic diagram of an LLM configuration tuning process provided in this public disclosure, combined with Figures 2 to 4 ,right Figure 1 The dynamic optimization method for hybrid solid-state drives based on large language models is further described.

[0038] This invention focuses on constructing a dynamic management framework for hybrid solid-state drives (SSDs) based on Large Language Models (LLMs), aiming to address the shortcomings of traditional storage management strategies in terms of globality and real-time performance in high-dimensional parameter space optimization. By systematically reconstructing the parameter system of hybrid SSDs and combining it with the contextual reasoning capabilities of LLMs, a novel scheme supporting adaptive tuning is proposed, overcoming the limitations of traditional machine learning models in parameter correlation mining and performance prediction accuracy.

[0039] The first step is a structural restructuring of the hybrid SSD parameter system. Hybrid SSD optimization involves multi-dimensional parameter coordination, including three core dimensions: hardware characteristic parameters, workload characteristic parameters, and policy control parameters. (1) Hardware characteristic parameters cover key configurations related to flash memory media, such as garbage collection (GC) trigger threshold, wear leveling strategy, space conversion granularity, etc., which directly affect the lifespan and performance of storage media. For example, an excessively high GC threshold will lead to the accumulation of invalid data and increase latency, while an excessively low threshold will exacerbate write amplification (WA).

[0040] (2) The workload characteristic parameters reflect the impact of data access patterns on storage strategies, including data hot classification cycle, read-write ratio, I / O burst intensity and other indicators. These need to be dynamically adjusted according to the load to optimize cache hit rate and energy efficiency.

[0041] (3) Policy control parameters are the control parameters of machine learning models and reinforcement learning (RL) algorithms, such as the convergence threshold of clustering algorithms and RL training intervals. Their rationality is directly related to the adaptability and stability of the policy.

[0042] By constructing a parameter sensitivity matrix, the study revealed significant differences in the impact of parameters on performance under varying workloads. In high random write scenarios, adjustments to the GC threshold and slice size had the most significant impact on latency and wait-and-see (WA); while in sequential read scenarios, the optimization weight for hot data classification cycles was higher. This finding provides important guidance for the attention mechanism of LLMs—the model needs to dynamically focus on the parameter combinations sensitive to the current workload and establish inference paths between parameters. For example, when the GC threshold is increased multiple times consecutively, the model needs to adjust the slice size in tandem to avoid potential performance degradation. This cross-parameter collaborative optimization capability transcends the expressive scope of traditional rule engines and shallow neural networks.

[0043] Secondly, this invention proposes a closed-loop optimization framework called LLM-SSD to execute the above-mentioned dynamic optimization method for hybrid solid-state drives based on large language models. This optimization framework mainly includes three core modules: a workload monitor, a thermal classification and space conversion module, and an LLM-driven automatic tuner.

[0044] (1) The workload monitor collects data such as I / O mode, flash memory health status, and cache hit rate in real time to generate a structured system status representation. Through a sliding window mechanism, the monitor can dynamically analyze workload volatility. Specifically, the standard deviation (SD) of the sliding window is used to quantify workload variability, and the calculation formula is as follows: .

[0045] (2) The thermal classification and spatial transformation module trains machine learning models based on historical data, such as the Hybrid Attention Model (HAML) to classify hot data, and dynamically adjusts the SLC / QLC region allocation through reinforcement learning strategies, such as Reinforcement Learning for conventional SolidState Drive (RL-cSSD).

[0046] (3) The LLM-driven automatic tuner receives system status and historical configuration information, generates new parameter configuration schemes, and ensures the effectiveness of the adjustment through a short-term performance verification module.

[0047] The innovation of this invention is also reflected in the design of the Prompt Engineering. By defining the LLM role as a "storage system expert," it is required to provide optimization suggestions based on input system information (such as current configuration, performance metrics, and workload statistics). Context enhancement strategies guide the model to generate a structured configuration list by embedding detailed background knowledge (such as parameter operating principles, historical tuning cases, and constraints). For example, when a high random write load is detected, the model might suggest "increasing the GC threshold from 6 to 7 to reduce WA, while adjusting the slice size to 16KB to balance latency and metadata overhead." The natural language parsing module then converts the text instructions into hardware-recognizable configuration files, such as JavaScript Object Notation (JSON), and filters invalid configurations through format validation and numerical range constraints. An error correction mechanism establishes an anomaly detection model that automatically rolls back to the previous stable version and re-triggers the tuning process when a new configuration causes performance degradation, ensuring system robustness.

[0048] The prompting phase is used to provide the large language model with detailed and accurate background information to generate effective responses. The prompts are designed in five key phases: First, role assignment, designating the large language model as a solid-state drive (SSD) expert; then, in the hybrid SSD overview and SSD management phase, providing the large language model with system information, including inherent configurations and current optimization strategies; in the historical configuration and performance phase, including a detailed list of past configuration changes, the reasons for the changes, and their performance impact; finally, clarifying requirements by defining a clear output format to ensure the large language model generates a more accurate list of new configurations.

[0049] The contextual modeling capabilities of LLMs play a crucial role in this invention. By maintaining a knowledge base containing thousands of historical tuning records, the model can capture the long-term effects and potential risks of parameter adjustments. For example, when the GC trigger threshold is increased multiple times consecutively, the model correlates this with potential write amplification risks and suggests adjusting the slice size synchronously to balance performance and lifetime. This cross-parameter associative reasoning capability stems from the advantages of LLMs in modeling sequential data, enabling them to identify temporal dependencies in parameter adjustments. Furthermore, the dynamic adjustment mechanism of the temperature parameter further enhances the model's adaptability: low-temperature settings (e.g., 0.7) encourage the model to adopt conservative strategies to maintain stability, while high-temperature settings (e.g., 1.2) stimulate exploratory behavior to cope with load fluctuations. Dynamically adjusting the temperature parameter effectively reduces the standard deviation of execution time, significantly improving the system's dynamic adaptability.

[0050] The detailed workflow is as follows: (1) First, determine the initial configuration set: using the default configuration of each optimization scheme and existing commercial solid-state drives, form the initial configuration file for the entire hybrid solid-state drive optimization framework. Next, use a parameter selection scheme to filter multiple parameters in the hybrid solid-state drive, and finally determine the configuration file to be input to the automatic tuner based on a large language model.

[0051] (2) Generating a new configuration file: Using the current configuration file, a new configuration file can be generated by the large language model-based auto-tuner at fixed intervals. The system monitor obtains system information, including current space occupancy, SLC mode ratio, and garbage collection count. The performance detector measures the performance for that period, taking into account average operating latency and write amplification. The new hybrid solid-state drive dynamic management framework (Large Language Model for Hybrid SolidState Drive, LLM-hybridSSD) uses historical configurations, system information, and performance data provided by the performance detector to initialize the large language model-based auto-tuner.

[0052] (3) Verifying the efficiency of the exploration configuration: After receiving the updated configuration list from the auto tuner, LLM-hybridSSD considers the next 10,000 operations as an observation period. The performance verification module calculates the average performance for the corresponding period and compares it with the performance of the previous period. If the performance drops by more than 5%, the configuration of the previous period is restored, and the auto tuner reacts to this phenomenon to obtain a new configuration.

[0053] In summary, the technological breakthroughs of this invention are reflected in three aspects: scalability of the parameter space, transparency of the optimization process, and improved energy efficiency. Through a hierarchical classification mechanism, the framework can seamlessly integrate new flash memory types (such as MLC) and emerging management strategies, such as the Compressed RAM Disk (ZRAM) caching layer, providing flexible scalability for future storage architecture evolution. The transparent interactive interface allows non-expert users to obtain customized configuration solutions by describing their needs in natural language (such as "reduce cold data access latency"), significantly lowering the barrier to entry for storage system tuning.

[0054] The following provides a detailed implementation process of the present invention, and further elaborates on the proposed LLM dynamic parameter tuning strategy and LLM error correction strategy in conjunction with the algorithm pseudocode.

[0055] 1. LLM dynamic parameter tuning strategy As shown in Table 1, the LLM dynamic parameter tuning strategy proposed in this invention utilizes LLMs to analyze the system status in real time (such as GC overhead and space utilization) and dynamically adjust the SLC / MLC / QLC conversion threshold and garbage collection strategy.

[0056] Table 1

[0057] 2. LLM Error Correction Strategy As shown in Table 2, the LLM error correction strategy proposed in this invention ensures that the configuration generated by LLM is legal and stable, and avoids system anomalies caused by inference errors.

[0058] Table 2

[0059] The innovativeness of this invention can be specifically reflected in the following aspects: (1) Systematic classification and optimization framework for hybrid SSD parameter space: For the first time, we classified the parameters of hybrid SSDs into SSD-specific parameters, workload-related parameters, and policy-related parameters, and identified key parameters that have a significant impact on performance. This systematic parameter classification and optimization framework provides clear guidance for the dynamic management of hybrid SSDs, enabling LLM to optimize within a more precise parameter space and avoiding performance bottlenecks caused by blind tuning.

[0060] (2) Proposal of the LLM-hybridSSD framework: We propose a novel framework, LLM-hybridSSD, which integrates LLM for dynamic tuning and configuration of hybrid solid-state drives. This framework fully utilizes the context-aware and optimization capabilities of LLM, transforming the complex hybrid SSD management problem into a dynamic tuning process, significantly improving the system's response time and write amplification performance.

[0061] (3) Application of LLM in Hybrid SSD Parameter Tuning. Through carefully designed Prompt Engineering, we transform the system status and historical configuration information of the hybrid SSD into natural language descriptions that LLM can understand, thereby guiding LLM to generate optimized parameter configurations. In addition, we developed a configuration generation and error correction mechanism to ensure that the configuration output by LLM is not only accurate but also effective.

[0062] This disclosure provides a method, apparatus, device, and storage medium for dynamic optimization of hybrid solid-state drives based on a large language model. It systematically classifies the parameter space of hybrid SSDs into three key parameters: hardware characteristics, workload features, and policy control. Utilizing the contextual reasoning and sequence modeling capabilities of the large language model, it constructs a closed-loop optimization framework that includes workload monitoring, thermal classification and space transformation, LLM automatic tuning, and performance verification. Through carefully designed prompting engineering, it transforms system states and historical configurations into natural language descriptions, guiding LLM to dynamically generate cross-parameter collaborative optimization configurations (such as linked adjustments to GC thresholds and slice sizes). Simultaneously, it introduces a dynamic temperature parameter adjustment mechanism to balance exploratory and conservative strategies, and configures error correction and performance rollback mechanisms to ensure system robustness. The technical advantages of this solution are as follows: Compared with traditional machine learning methods that only optimize a single subset and lack the ability to correlate global parameters, this invention can sense workload changes in real time and dynamically focus on the most sensitive parameter combination under the current load, realizing collaborative reasoning and adaptive adjustment among multiple parameters. This significantly reduces write amplification and operation latency, improves the response efficiency, system stability and energy consumption of hybrid SSDs under dynamic workloads, and its parameter hierarchical classification mechanism also provides a scalable architectural foundation for the integration of new flash memory types and management strategies in the future.

[0063] and Figure 1 Corresponding to the dynamic optimization method for hybrid solid-state drives based on large language models disclosed in the previous disclosure, this disclosure also provides a dynamic optimization device for hybrid solid-state drives based on large language models, such as... Figure 5 As shown, the hybrid solid-state drive dynamic optimization device based on a large language model may include: The classification module 501 is used to systematically classify the parameter space of the hybrid solid-state drive into hardware characteristic parameters, workload characteristic parameters and policy control parameters, and construct a parameter sensitivity matrix based on the classified parameters to determine the difference in the impact of parameters on performance under different workloads. The acquisition module 502 is used to acquire system status information of the hybrid solid-state drive in real time, including I / O mode, flash health status, cache hit rate, space utilization, SLC mode ratio, garbage collection count, average operation latency and write amplification. The optimization module 503 is used to generate an optimized parameter configuration scheme based on the large language model, the system status information, the influencing difference parameters, and the historical configuration information.

[0064] In some embodiments, the optimization module 503 can also be used to perform performance verification on the parameter configuration scheme. If the performance improvement is lower than a preset threshold or the performance decline exceeds a preset ratio, the configuration will be rolled back to the previous configuration.

[0065] In some embodiments, the above hardware characteristic parameters include garbage collection trigger threshold, wear leveling strategy, and space conversion granularity; Workload characteristics include data popularity classification period, read / write ratio, and I / O burst intensity; Policy control parameters include the clustering algorithm convergence threshold and the reinforcement learning training interval.

[0066] In some embodiments, the above system status information is collected by a workload monitor, which uses a sliding window mechanism to calculate the variability of the workload. The standard deviation is calculated using the following formula: , in For window size, For the first One observation value, This represents the mean value within the window.

[0067] In some embodiments, when generating an optimized parameter configuration scheme based on a large language model, a dynamic temperature parameter adjustment mechanism is introduced; When the workload variability is below the first threshold, the temperature parameter is set to a preset low temperature range, so that the large language model is biased to output a conservative strategy similar to the historical stable configuration. When the workload variability exceeds the second threshold, the temperature parameter is set to a preset high temperature range to increase the randomness and exploratory nature of the output configuration of the large language model in order to adapt to sudden load changes.

[0068] In some embodiments, the above-described optimized parameter configuration scheme includes: The large language model outputs a structured configuration list, which is then converted into a hardware-recognizable configuration file by the natural language parsing module, and subjected to format validation and numerical range constraint filtering.

[0069] In some embodiments, the optimization module 503 can also be used to take the preset time after generating the configuration as the observation period, calculate the average performance during the corresponding period and compare it with the previous period. If the performance drops by more than 5%, the previous configuration will be restored and configuration generation will be triggered again.

[0070] Combination Figure 6As shown in the illustration, this disclosure also provides a hybrid solid-state drive dynamic optimization device 600 based on a large language model, including a processor 604 and a memory 601. Optionally, the system may further include a communication interface 602 and a bus 603. The processor 604, communication interface 602, and memory 601 can communicate with each other via the bus 603. The communication interface 602 can be used for information transmission. The processor 604 can call logical instructions in the memory 601 to execute the hybrid solid-state drive dynamic optimization method based on a large language model described in the above embodiments.

[0071] Furthermore, the logic instructions in the aforementioned memory 601 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0072] The memory 601, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 604 executes functional applications and data processing by running the program instructions / modules stored in the memory 601, thereby realizing the hybrid solid-state drive dynamic optimization method based on a large language model in the above embodiments.

[0073] The memory 601 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 601 may include high-speed random access memory and may also include non-volatile memory.

[0074] This disclosure provides a computer-readable storage medium storing computer-executable instructions configured to execute a hybrid solid-state drive dynamic optimization method based on a large language model.

[0075] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0076] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0077] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0078] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0079] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0080] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0081] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0082] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this disclosure can be achieved, and this is not limited herein.

[0083] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A dynamic optimization method for hybrid solid-state drives based on a large language model, characterized in that, The method includes: The parameter space of hybrid solid-state drives is systematically classified into hardware characteristic parameters, workload characteristic parameters, and policy control parameters. Based on the classified parameters, a parameter sensitivity matrix is ​​constructed to determine the differences in the impact of parameters on performance under different workloads. Real-time acquisition of system status information of hybrid solid-state drives, including I / O mode, flash memory health status, cache hit rate, space utilization, SLC mode ratio, garbage collection count, average operation latency, and write amplification rate; Based on the large language model, an optimized parameter configuration scheme is generated according to the system state information, influencing difference parameters, and historical configuration information.

2. The method according to claim 1, characterized in that, The method further includes: The performance of the parameter configuration scheme is verified. If the performance improvement is lower than the preset threshold or the performance decline exceeds the preset ratio, the previous configuration is rolled back.

3. The method according to claim 1, characterized in that, The hardware characteristic parameters include garbage collection trigger threshold, wear leveling strategy, and space conversion granularity; The workload characteristic parameters include data heat classification period, read-write ratio, and I / O burst intensity; The policy control parameters include the clustering algorithm convergence threshold and the reinforcement learning training interval.

4. The method according to claim 1, characterized in that, The system status information is collected by a workload monitor, which uses a sliding window mechanism to calculate the workload variability. The standard deviation is calculated using the following formula: ， in For window size, For the first One observation value, This represents the mean value within the window.

5. The method according to claim 1, characterized in that, When generating optimized parameter configuration schemes based on large language models, a dynamic temperature parameter adjustment mechanism is introduced. When the workload variability is below the first threshold, the temperature parameter is set to a preset low temperature range, so that the large language model is biased to output a conservative strategy similar to the historical stable configuration. When the workload variability exceeds the second threshold, the temperature parameter is set to a preset high temperature range to increase the randomness and exploratory nature of the output configuration of the large language model in order to adapt to sudden load changes.

6. The method according to claim 1, characterized in that, The optimized parameter configuration scheme includes: The large language model outputs a structured configuration list, which is then converted into a hardware-recognizable configuration file by the natural language parsing module, and subjected to format validation and numerical range constraint filtering.

7. The method according to claim 1, characterized in that, The method further includes: The preset time after generating the configuration is used as the observation period, and the average performance during the corresponding period is calculated and compared with the previous period. If the performance drops by more than 5%, the previous configuration will be restored and configuration generation will be triggered again.

8. A hybrid solid-state drive dynamic optimization device based on a large language model, characterized in that, For performing the method according to any one of claims 1 to 7; the apparatus comprises: The classification module is used to systematically classify the parameter space of hybrid solid-state drives into hardware characteristic parameters, workload characteristic parameters, and policy control parameters. Based on the classified parameters, a parameter sensitivity matrix is ​​constructed to determine the differences in the impact of parameters on performance under different workloads. The acquisition module is used to collect system status information of the hybrid solid-state drive in real time, including I / O mode, flash health status, cache hit rate, space utilization, SLC mode ratio, garbage collection count, average operation latency and write amplification. The optimization module is used to generate an optimized parameter configuration scheme based on the large language model, the system state information, the parameters affecting differences, and the historical configuration information.

9. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images