Multi-thread based solid state disk reliability automated testing method and system

Abstract

The present application relates to the technical field of hard disk testing and verification, in particular to a solid state disk reliability automatic testing method and system based on multi-threading, which can quickly capture the subtle performance fluctuations of the solid state disk under concurrent load and identify fault modes by means of microsecond-level high-granularity continuous performance monitoring and deep analysis of SMART information, supplemented by multi-thread parallel processing, then early warning is realized by using the extracted multi-dimensional features and machine learning model, and a detailed fault diagnosis report is generated; through dynamic error correction code strength verification and data integrity verification under pressure, the internal error correction mechanism of the solid state disk is actively detected and optimized. Finally, combined with the predicted reliability modeling, the system can estimate the remaining service life, predict the failure, and provide product optimization suggestions for the design, manufacturing and firmware optimization of the solid state disk, so as to realize the automation, intelligentization and full life cycle management of the solid state disk fault detection reliability.

Description

Technical Field

[0001] This invention relates to the field of hard disk testing and verification technology, specifically to an automated testing method and system for the reliability of solid-state drives based on multi-threading. Background Technology

[0002] SMART monitoring typically employs a passive response mechanism triggered by fixed thresholds. This post-event reporting mode cannot provide early warnings. By the time an error is detected, the optimal intervention time has already been missed, leading to a wide-ranging impact from subsequent failures.

[0003] Intermittent errors, often caused by power supply disturbances and signal interference, occur instantaneously during testing and are difficult to detect quickly using conventional methods. Logical errors such as data races and sequence disorders caused by concurrent operations result in blind spots in detection coverage. Current technologies typically only provide superficial error information, such as reporting read / write errors at a logical block address or an increase in the bad block count in SMART attributes. However, this superficial information lacks the ability to analyze the root cause of underlying faults, making it difficult to trace the root cause and fundamentally solve the problem. When an error is detected, engineers struggle to accurately distinguish and locate whether the root cause is physical media damage, controller failure, or firmware defect. The incompleteness of the testing and verification system is also a prominent issue, lacking a systematic fault injection platform and recovery verification mechanism, making it impossible to comprehensively assess the equipment's fault tolerance and data integrity assurance level under abnormal conditions.

[0004] To address this, a multi-threaded automated testing method and system for solid-state drive reliability is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-threaded automated testing method and system for solid-state drive (SSD) reliability, enabling early warning, accurate diagnosis, effective correction, and intelligent monitoring throughout the SSD's lifecycle for SSD failures. Specifically:

[0006] High-granularity continuous performance monitoring is adopted to continuously collect and record real-time performance parameters of solid-state drives under multi-threaded concurrency with microsecond-level time granularity; combined with the real-time performance parameters and deeply analyzed SMART information, instantaneous performance fluctuations are captured and fault modes are identified through multi-threaded parallel processing.

[0007] Extract the multi-dimensional features of the fault modes, historical performance indicators, real-time performance parameters, and health performance indicators; use machine learning models to detect subtle signs of performance degradation for early warning, and generate fault diagnosis reports with timestamps and associated performance and health data.

[0008] It performs dynamic error correction code strength verification and stress data integrity verification to proactively detect and verify the effectiveness of the internal error correction mechanism of the solid-state drive. At the same time, through predictive reliability modeling, it uses multi-source heterogeneous data to estimate the remaining service life and predict failures, and generates product optimization suggestions for reliability management.

[0009] Preferably, the real-time performance parameters include input / output operations per second, throughput, latency performance indicators, mixed read / write load, and background garbage collection and wear leveling simulation. The SMART information includes, but is not limited to, the original error count of the flash memory chip, bad block management information, wear leveling log, and internal firmware log. The SMART information is deeply analyzed to obtain the internal structure, thresholds, and correlations between various attributes.

[0010] Multiple independent parallel threads are configured to quickly capture minute instantaneous performance fluctuations of the solid-state drive under multi-threaded concurrent load by cross-comparing the real-time performance parameters with the deeply analyzed SMART information in real time, and to identify fault modes that deviate from normal behavior based on preset rules or preliminary models.

[0011] Preferably, the specific steps for extracting the multi-dimensional features include: extracting and constructing features covering the time domain, frequency domain, and statistical domain from the fault mode data, historical performance indicators, real-time performance parameters, and health performance indicators; filtering and dimensionality reduction of the features; and constructing the interaction and ratio between the performance indicators and health indicators.

[0012] Preferably, the use of a machine learning model to detect subtle signs of performance degradation and achieve early warning of the health status of the solid-state drive specifically includes: using a pre-trained machine learning model to learn and analyze the multi-dimensional features, detecting even subtle signs of performance degradation within the normal fluctuation range, and distinguishing them from occasional noise; when the abnormal score or classification result output by the machine learning model reaches a preset warning threshold, the system immediately triggers an early warning signal and automatically generates a fault diagnosis report, which has a precise timestamp and is associated with the specific performance indicators and health data that led to the warning trigger.

[0013] Preferably, the dynamic error correction code strength verification and stress-based data integrity verification method specifically involves: designing a series of adaptive, high-stress test scenarios to induce bit errors in NAND flash memory under simulated extreme temperature, voltage fluctuations, high write wear, or specific data patterns; the system monitors the memory's error correction capability in real time and dynamically adjusts at least one of the following operating strategies of the NAND flash memory: flash memory refresh cycle, write voltage, read reference voltage, bad block management algorithm, ECC algorithm strength switching, read retry mechanism, data refresh or data migration strategy, power consumption management strategy, and data scrambling strategy, and verifies the data recovery and integrity maintenance capabilities under extreme conditions.

[0014] Preferably, the predictive reliability modeling and remaining lifespan estimation method specifically involves: integrating multi-source heterogeneous data such as continuous performance data of the solid-state drive, detailed SMART health attributes, manufacturing batch information, historical test results, and actual field failure data; using the random forest algorithm to construct a full lifecycle reliability prediction model, estimating the remaining lifespan of the solid-state drive, and predicting potential failure time points and modes.

[0015] Preferably, the product optimization recommendations for generating reliability management specifically include: analyzing the reliability performance of the solid-state drive under different loads, environments, or batches based on the failure time point and the failure mode, identifying key influencing factors and degradation modes that lead to performance degradation or potential failures, and summarizing the analysis results; and generating specific product optimization recommendations for solid-state drive design, manufacturing process, firmware optimization, testing strategies, and preventive maintenance plans based on the analysis results.

[0016] An integrated automated reliability testing system for solid-state drives (SSDs) based on multi-threading is included, comprising the following modules:

[0017] The high-granularity performance data acquisition module is used to continuously collect and record real-time performance parameters of solid-state drives under multi-threaded concurrency with high-granularity continuous performance monitoring at the microsecond time granularity, including input / output operations per second, throughput, latency performance indicators, mixed read / write load, and background garbage collection and wear leveling simulation.

[0018] The data integration and preliminary anomaly identification module is used to combine the real-time performance parameters with the deeply analyzed SMART information, and capture instantaneous performance fluctuations and identify fault modes through multi-threaded parallel processing.

[0019] A multi-dimensional feature engineering module is used to extract the fault modes, historical performance indicators, real-time performance parameters, and health performance indicators to perform multi-dimensional feature extraction.

[0020] The fault warning and anomaly report generation module uses machine learning models to detect subtle signs of performance degradation for early warning and generates fault diagnosis reports with timestamps, associated performance and health data.

[0021] The reliability verification and lifetime prediction module is used to perform dynamic error correction code strength verification and data integrity verification under stress, actively probing and verifying the effectiveness of the internal error correction mechanism of the solid-state drive; at the same time, through predictive reliability modeling, it uses multi-source heterogeneous data to estimate the remaining lifetime and predict failures, generating product optimization suggestions for reliability management.

[0022] Compared with the prior art, the technical effects of the present invention are as follows:

[0023] 1. This invention combines high-granularity continuous performance monitoring with deep analysis of SMART information, and employs multi-threaded parallel processing technology to capture subtle performance fluctuations and transient anomalies of solid-state drives under concurrent loads. Furthermore, by combining extracted multi-dimensional features and machine learning models, it can detect performance degradation signs that are difficult to detect using traditional methods and provide early warnings. This allows the system to intervene before faults escalate, shifting from passive response to proactive prevention, greatly improving the sensitivity of error detection and the accuracy of diagnosis, thereby effectively avoiding data loss and system downtime.

[0024] 2. This method innovatively introduces dynamic error correction code strength verification and stress-based data integrity verification. It can proactively simulate and verify the effectiveness of the internal error correction mechanisms of solid-state drives (SSDs) under extreme environments, such as power disturbances and high write wear, including ECC algorithms, bad block management, and data refresh strategies. By verifying the data recovery and integrity maintenance capabilities of SSDs under extreme conditions, it ensures that SSDs can maintain normal operation under complex and abnormal operating conditions, preserving data integrity and reducing information loss caused by sudden events.

[0025] 3. This invention integrates multi-source heterogeneous data, including performance, health, manufacturing, and historical failure data, and uses predictive algorithms to estimate the remaining lifespan and predict failures of solid-state drives (SSDs). It not only provides intuitive health status monitoring, but more importantly, it can analyze failure modes and influencing factors based on the prediction results, thereby generating product optimization suggestions for design, manufacturing, firmware optimization, and preventative maintenance. This forward-looking, full-lifecycle intelligent monitoring and decision support is a crucial component of SSD reliability testing, significantly improving the reliability of SSD fault diagnosis and prediction. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the automated reliability testing method for solid-state drives based on multi-threading proposed in this application.

[0027] Figure 2 This is a schematic diagram of the automated solid-state drive reliability testing system based on multi-threading proposed in this application.

[0028] Figure 3 This is a schematic diagram of the adaptive verification and reinforcement learning lifetime prediction process proposed in this application. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figures 1 to 3 This invention provides an automated reliability testing method and system for solid-state drives based on multi-threading. The technical solution is as follows: (Refer to...) Figure 1 This is a flowchart illustrating the automated reliability testing method for solid-state drives based on multi-threading proposed in this application, specifically including:

[0031] High-granularity continuous performance monitoring is adopted to continuously collect and record real-time performance parameters of solid-state drives under multi-threaded concurrency with microsecond-level time granularity; combined with the real-time performance parameters and deeply analyzed SMART information, instantaneous performance fluctuations are captured and fault modes are identified through multi-threaded parallel processing.

[0032] Multi-dimensional feature extraction is performed on the fault modes, historical performance indicators, real-time performance parameters, and health performance indicators; machine learning models are used to detect subtle signs of performance degradation for early warning, and fault diagnosis reports with timestamps and associated performance and health data are generated.

[0033] It performs dynamic error correction code strength verification and stress data integrity verification to proactively detect and verify the effectiveness of the internal error correction mechanism of the solid-state drive. At the same time, through predictive reliability modeling, it uses multi-source heterogeneous data to estimate the remaining service life and predict failures, and generates product optimization suggestions for reliability management.

[0034] Integrate a system, refer to Figure 2 This is a schematic diagram of the structure of the automated solid-state drive reliability testing system based on multi-threading proposed in this application.

[0035] Example 1: This example provides a multi-threaded automated reliability testing method for solid-state drives (SSDs). Through high-granularity performance monitoring, intelligent data analysis, and machine learning prediction, it enables early warning of SSD health status, fault diagnosis, and product optimization suggestions. It overcomes the limitations of traditional SSD testing methods, particularly in capturing instantaneous performance fluctuations and predicting potential failures, providing strong technical support for SSD R&D, production, and quality control.

[0036] The real-time performance parameters include IOPS (Input / Output Operations Per Second), throughput, latency performance metrics, mixed read / write loads, and background garbage collection and wear leveling simulations. Sub-microsecond performance monitoring based on hardware interrupt capture and a custom high-speed counter, supplemented by event tracking and analysis methods, achieves ultra-high-precision real-time acquisition of parameters such as latency, queue depth, and command completion time for each key node on the SSD I / O path. This method can capture nanosecond-level and even picosecond-level micro-fluctuations in performance that are difficult to detect with traditional software-level monitoring, and construct a "behavioral fingerprint" dataset of the SSD under normal and abnormal states. Furthermore, the real-time data stream shows that IOPS fluctuated from 100,000 to 50,000 under write-intensive loads, throughput plummeted from 2GB / s to 0.5GB / s, and the average read latency spiked from 100μs to 500μs. Simultaneously, real-time analysis of the SMART information stream revealed that the raw error count suddenly increased from 0 to 50 per hour, the number of bad blocks increased from 0 to 3 within 24 hours, and the wear leveling count plummeted from 90% to 85% within 1 hour.

[0037] Multiple independent parallel threads are configured to analyze real-time performance parameters and deeply analyzed SMART information. Specifically, deep analysis involves establishing a SMART attribute association model based on a graph neural network to learn and identify complex nonlinear relationships between attributes, such as the multi-path dependency between bad block count and read / write error rate. An adaptive multi-agent system is deployed, with each agent responsible for analyzing different categories of SMART attributes (e.g., raw error counting agent, wear leveling agent, temperature agent, etc.). These agents collaborate through a distributed consensus mechanism to analyze the temporal evolution and cross-attribute associations of SMART attributes, as well as multimodal data fusion with external environmental parameters. This significantly improves the accuracy and robustness of SSD fault diagnosis, enabling earlier warnings and more accurate root cause analysis, while also considering adaptability across different devices and data privacy protection thresholds. Historical baseline data and the statistical anomaly detection method IQR are acquired, and the analysis model and warning thresholds are dynamically adjusted based on the SSD's NAND type, controller model, and firmware version. The warning thresholds for each SMART attribute are also dynamically adjusted to adapt to the normal fluctuation range of the SSD at different lifecycle stages. Internal structure analysis refers to identifying the physical or logical meaning behind SMART attributes. For example, it involves real-time cross-comparison of the correspondence between raw error counts and specific NAND flash memory pages or blocks, quickly capturing minute instantaneous performance fluctuations of solid-state drives under multi-threaded concurrent loads, and identifying fault modes that deviate from normal behavior based on preset rules or preliminary models.

[0038] By monitoring and analyzing the operating metrics of solid-state drives (SSDs) in real time and performing real-time analysis, and combining the interaction between multiple operating metrics, the sensitivity and speed of SSD fault identification can be significantly improved. Especially in high-concurrency and complex load scenarios, it can detect subtle performance degradation and potential faults that are difficult to detect by traditional methods as early as possible.

[0039] Furthermore, the specific steps for extracting multi-dimensional features include:

[0040] Features covering the time, frequency, and statistical domains are extracted and constructed from the fault mode data, historical performance metrics, real-time performance parameters, and health performance metrics. Specifically, time-domain features include an average of 95,000 IOPS over 10 consecutive seconds and a delay standard deviation of 50 μs over 5 minutes; frequency-domain features include a major frequency component with an abnormally high amplitude greater than 1.5 relative units at 0.5 Hz after Fourier transform of the IOPS time series; statistical domain features mainly involve descriptive analysis of the acquired data, including a 99th percentile of read latency of 250 μs and a change rate of error count of 0.05 errors / second. Furthermore, statistical features such as kurtosis, skewness, and cross-correlation within a sliding window can be extracted to examine the data's distribution pattern and skewed distribution. These features are then filtered and dimensionality-reduced; for example, PCA is used to reduce the dimensionality of the 50 original features to 10 principal components, retaining 98% of the information. The interaction and ratios between performance metrics and health metrics can be constructed. For example, the ratio of throughput to the number of bad blocks, or the product of average latency and wear leveling progress, can be calculated. Alternatively, the ratio between the rate of change of the original error count and the total number of correctable errors in the SMART attribute can be constructed. Combining workload with these metrics can standardize error indicators, preventing the natural increase in errors under high load from being misjudged as a failure, and helping to distinguish between normal wear and tear due to heavy workload and inherent hardware defects.

[0041] By constructing a multi-dimensional, highly cohesive feature set, richer and more comprehensive data support is provided for the accurate learning of subsequent machine learning models. Then, principal component analysis is used to perform sufficient dimensionality reduction, and the most influential feature combinations are selected as inputs to the machine learning model, which effectively improves the accuracy of fault diagnosis and the generalization ability of the model.

[0042] Furthermore, the use of machine learning models to detect subtle signs of performance degradation, enabling early warning of the solid-state drive's health status, specifically includes:

[0043] A pre-trained support vector machine (SVM) model was used, employing a Gaussian radial basis function as the kernel function, with the C parameter set to 10 and the gamma parameter set to 0.1. The training dataset contained 10,000 samples, with 80% representing normal operation data and 20% representing data from different failure modes. The data underwent rigorous cleaning, timestamp alignment, and Min-Max normalization. Hyperparameter tuning was performed using grid search, and 5-fold cross-validation was used to evaluate model performance. The multi-dimensional features were learned and analyzed to detect even subtle performance degradation signs within normal fluctuation ranges, such as a 0.5% continuous decrease in IOPS even when the IOPS fluctuated within the normal range of 5%, distinguishing it from occasional noise. Furthermore, in 1000 simulation tests, the model achieved a 95% accuracy rate in detecting a 0.5% continuous performance degradation, with a false positive rate of less than 1%. When the anomaly score output by the SVM model exceeds 0.8 or is classified as a "potential failure," reaching a preset warning threshold, the system immediately triggers an early warning signal and automatically generates a fault diagnosis report. The report includes a precise timestamp and is linked to specific performance metrics and health data that triggered the alert, such as an abnormal increase in the 99th percentile latency to 300μ and a raw error count that increased by 20 in one hour.

[0044] By using machine learning models for pattern classification, subtle signs of potential SSD failures can be identified, enabling ultra-early warnings of SSD health status. This allows for intervention before failures occur, significantly reducing unexpected downtime and providing accurate diagnostic data for after-sales maintenance.

[0045] Furthermore, the dynamic error correction code strength verification and stress-based data integrity verification method specifically includes:

[0046] A series of adaptive, high-stress test scenarios were designed to simulate extreme temperatures, such as 85°C for 24 hours, voltage fluctuations, such as the supply voltage VCC fluctuating from 3.3V to 2.97V and then to 3.63V at 5 times / second, high write wear, such as writing 100TB per day for 7 consecutive days, with a total write volume of 700TB, or specific data pattern conditions to induce bit errors in NAND flash memory, such as inducing 1000 bit flips in a 1TB data block. For NAND flash memory, there are many data patterns that can lead to bit errors. The specific data patterns include checkerboard pattern, reverse repeating pattern, all 0 or all 1 pattern, random data pattern, and specific programming sequences that maximize the coupling interference of neighboring cells. The system monitors the error correction capability of the memory in real time, captures and analyzes high-granularity indicators such as the number of original error correction code check bits, the number of uncorrected errors, and the data reconstruction success rate exposed in the underlying NAND controller log, and evaluates the error correction margin and stability of the current memory of the solid-state drive in real time through a dynamic risk assessment model based on Bayesian networks, and identifies potential error correction failure trends. For example, after each data write, the number of ECC check bits may remain in a low, stable range (e.g., 2-3 bits per check), the uncorrected error count remains 0, and the data reconstruction success rate remains 100%. When the solid-state drive (SSD) begins to age or encounters write pressure, the number of ECC check bits may gradually increase to (e.g.) 5-8 bits, while the uncorrected error count occasionally shows non-zero values. This trend is then identified, and combined with historical data, it is determined that although the data can still be corrected, the error correction margin is decreasing, providing an early warning of a "potential error correction failure trend." For example, it is predicted that the frequency of uncorrected error counts may exceed a certain threshold within the next 100 hours, indicating a risk of data loss. A dynamic threshold is set, such as triggering a strategy adjustment when the average number of corrected bits per megabyte exceeds 100. Furthermore, at least one of the following operating strategies of the NAND flash memory is dynamically adjusted:

[0047] The flash refresh cycle has been shortened from 1 hour to 30 minutes (triggered when the read error rate of a NAND block exceeds 0.1% within 10 consecutive minutes), the write voltage has been dynamically increased from 1.8V to 2.0V (incrementing in 0.05V steps when the media error rate increases by 5% within 1 hour), the read reference voltage and bad block management algorithm have been changed from static mapping to dynamic remapping (triggered when the number of bad blocks increases by more than 3 within 24 hours), the ECC algorithm strength has been changed from LDPC 8K to LDPC 16K (switching immediately when the number of uncorrected errors occurs within 1 second), the read retry mechanism has been increased from 1 retry to 3 retryes (retry is initiated when the first read fails), the data refresh or data migration strategy actively migrates high error rate data blocks to healthy data blocks (triggered when the original error count of a block exceeds the threshold of 5000), power consumption management strategy, and data scrambling strategy have been implemented, and the data recovery and integrity maintenance capabilities under extreme conditions have been verified.

[0048] Reference Figure 3 This diagram illustrates the adaptive verification and reinforcement learning lifetime prediction process proposed in this application. Specifically, the verification criterion is set to ensure a 100% data recovery rate after inducing 1000 bit errors by reading and comparing the original data. Further evaluation of the recovery time ensures data recovery is completed within 25 milliseconds, and the impact on normal IOPS performance during recovery is quantified. For example, if the performance degradation does not exceed 5%, data integrity and acceptable service quality are ensured.

[0049] Furthermore, at least one of the following operating strategies of the NAND flash memory is dynamically adjusted: the flash refresh cycle is shortened from 1 hour to 30 minutes, the write voltage is dynamically increased from 1.8V to 2.0V, the read reference voltage is adjusted, the bad block management algorithm is switched from static mapping to dynamic remapping, the ECC algorithm strength is switched from LDPC 8K to LDPC 16K, the read retry mechanism is increased from 1 retry to 3 retryes, the data refresh or data migration strategy is to actively migrate high error rate data blocks to healthy data blocks, the power consumption management strategy is adjusted, and the data scrambling strategy is adjusted. The data recovery and integrity maintenance capabilities under extreme conditions are verified to ensure that the data recovery rate is still 100% after inducing 1000 bit errors.

[0050] By verifying the adaptive capabilities of data integrity and error correction code verification under harsh conditions, the solid-state drive (SSD) can still work reliably and protect user data under extreme conditions, preventing large-scale data loss due to sudden events. This greatly enhances the product's robustness and reduces the data loss rate of the SSD within its expected lifespan.

[0051] Furthermore, the predictive reliability modeling and remaining useful life estimation method specifically includes:

[0052] This process integrates continuous performance data from solid-state drives (SSDs), such as daily average IOPS, latency, 99.9% latency quantile, detailed SMART health attributes like NAND Life Left percentage, Power-On Hours, Host Writes, manufacturing batch information, historical test results (e.g., 1000 hours of trouble-free factory aging tests), and real-world failure data (e.g., 500 failure reports from 10,000 SSDs over the past 5 years), among other heterogeneous data from multiple sources. The data integration process includes: aligning data from different sources using a unified timestamp; completing missing data using multiple imputation algorithms; performing one-hot encoding on categorical data; performing Z-score normalization on continuous numerical data; and applying different data filtering, denoising, or standardization strategies based on the SSD's current operating mode (e.g., read-intensive, write-intensive, idle, and real-time health status) to ensure that data preprocessing accurately reflects the device's real-world behavior in specific scenarios.

[0053] Furthermore, a full lifecycle reliability prediction model was constructed using the random forest algorithm. This model was trained on data from 10,000 solid-state drives (SSDs), comprising 500 decision trees, each with a maximum depth of 15 and a minimum leaf node sample size of 5. Feature random sampling was employed to enhance the model's generalization ability. The model estimates the remaining lifespan of the SSDs in days using regression analysis and predicts potential failure modes, such as NAND wear, controller failure, and firmware anomalies, through a multi-class classification task, providing corresponding prediction probabilities or confidence levels. When predicting the remaining lifespan, the mean absolute error was less than 30 days, and the prediction accuracy for potential failures reached 92%. The model estimates the remaining lifespan of the SSDs, such as predicting SSD#1001's remaining lifespan to be 450 days, and predicts potential failure times, such as September 15, 2026, and patterns, such as read errors caused by NAND wear.

[0054] It provides predictions of the remaining lifespan of solid-state drives (SSDs) and potential failures, which is crucial for data centers or enterprise users to conduct proactive maintenance and resource planning. It can effectively reduce the risk of data loss and operating costs, and through preventative maintenance, effectively reduce the maintenance costs of SSDs.

[0055] Furthermore, the product optimization suggestions for generating reliability management specifically include:

[0056] Based on the estimated remaining lifespan and failure predictions, a random forest algorithm was used to analyze the reliability performance of solid-state drives (SSDs) under different loads, environments, or batches. Key influencing factors and degradation patterns leading to performance degradation or potential failures were identified, and the analysis results were summarized. Specifically, the analysis results may reveal that for batch A123 SSDs, the probability of failure within one year significantly increases from 1% to 5% when continuous read / write loads exceed 200GB / day. The main failure mode is excessive write amplification leading to accelerated NAND wear. The summarized analysis results found that 70% of early failures are related to poor garbage collection efficiency of the firmware under specific mixed loads. Based on the analysis results, specific product optimization suggestions were generated for solid-state drives (SSDs), such as increasing the proportion of spare blocks from 7% to 10%, improving manufacturing processes (e.g., optimizing NAND flash memory selection standards and increasing the minimum P / E cycle count from 3000 to 4000), firmware optimization (e.g., updating the garbage collection algorithm to improve efficiency by 15% at 80% disk fullness), testing strategies (e.g., adding 1000 hours of mixed read / write stress testing), and preventative maintenance plans (e.g., recommending users perform data migration when SSD health is below 20%).

[0057] Lifespan estimation can provide data-driven optimization directions for the design, manufacturing, and maintenance of solid-state drives (SSDs), allowing users to copy important data in advance, providing avoidance methods to extend the lifespan of the drive, thereby continuously improving the overall reliability of the product, reducing the probability of future SSD failures, and extending the service life.

[0058] This embodiment presents a multi-threaded automated solid-state drive (SSD) reliability testing method. By integrating real-time performance monitoring, deep SMART information parsing, multi-dimensional feature engineering, advanced machine learning early warning, dynamic error correction verification, and predictive reliability modeling, it achieves comprehensive, accurate, and proactive management of SSD health status. It can detect minute signs of performance degradation and potential faults at very early stages, helping users to perform precise preventative maintenance and resource planning. This invention consists of a highly integrated intelligent system that works collaboratively in a multi-threaded parallel architecture, achieving full automation from data acquisition to decision optimization.

[0059] Example 2: This example builds upon existing methods and is specifically designed to address the challenges of high concurrency, mixed read / write operations, low latency, and complex data patterns faced by solid-state drives (SSDs) in edge computing and artificial intelligence applications. This example emphasizes ultra-high granularity I / O behavior capture and cross-monitoring of system resources. It collects real-time data on the SSD's input / output operations per second, throughput, and granular latency at 128 queue depths. Simultaneously, mixed data patterns and load simulations encompass sensor data streams, real-time inference data, and model updates. Crucially, it simultaneously collects the CPU utilization and memory usage of edge devices to differentiate between SSD-related issues and system resource bottlenecks.

[0060] Furthermore, data integration is strengthened by incorporating application-layer context information. By deeply correlating intelligent monitoring information with the internal firmware event logs of the solid-state drive (SSD), and receiving real-time information such as task type, data processing stage, and model version from edge computing applications, the source of problems can be accurately diagnosed. Combining this data with complex pattern matching rules, up to 50 fault modes can be quickly identified and categorized, such as "SSD read amplification caused by model loading."

[0061] In addition to conventional statistical features, advanced time-series features such as autocorrelation, cross-correlation, and wavelet transform coefficients of performance indicators are extracted to capture the periodicity and abrupt changes in performance. Simultaneously, external environmental sensor data such as temperature, humidity, and vibration are incorporated to construct the interaction between solid-state drive performance and environmental factors, revealing the impact of the environment on reliability.

[0062] This embodiment employs a Temporal Convolutional Neural Network (TCN) model for fault early warning. This deep learning network excels at processing time-series data and can identify subtle signs of performance degradation, even within normal fluctuation ranges, where patterns have changed. An attention mechanism is incorporated into the model, enabling the early warning report to highlight features that contribute most to the judgment, providing stronger fault interpretation capabilities. Multiple early warning thresholds are set to generate tiered diagnostic reports and provide hierarchical suggestions.

[0063] Furthermore, based on the real-time health status of the solid-state drive (SSD), the system dynamically generates customized error-induced test scenarios and adaptively adjusts strategies such as error correction code algorithm strength and flash memory refresh cycle to verify data integrity under extreme conditions. Simultaneously, a lifetime prediction model is built using reinforcement learning algorithms. Through interaction with a simulated environment, it learns how to more accurately estimate remaining lifespan under different operating conditions and aging stages, providing data-driven support for product optimization.

[0064] The optimization suggestions in this embodiment are more targeted. Based on all collected and analyzed data, the system performs in-depth root cause analysis of faults and, combined with a predefined knowledge graph, automatically generates highly specific optimization suggestions. Furthermore, considering the characteristics of edge computing devices, it provides optimization suggestions for solid-state drive selection, configuration, and maintenance, including heat dissipation improvements, firmware policy adjustments, or energy-saving optimizations. Finally, based on accurate lifespan predictions, it automatically generates detailed preventative maintenance and firmware update plans.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automated testing of reliability of a multi-thread based solid state drive, characterized in that, The application relates to a solid state disk health management method and system. High-granularity continuous performance monitoring is adopted to continuously collect and record real-time performance parameters of a solid state disk under multi-thread concurrency at a microsecond time granularity; In combination with the real-time performance parameters and deep analysis of SMART information, a multi-thread parallel processing is adopted to capture performance transient fluctuation and identify fault modes; the SMART information is deeply analyzed to obtain internal structures, threshold values and correlations between attributes; A plurality of independent parallel threads are configured to quickly capture slight performance transient fluctuation through real-time cross comparison and preliminarily identify fault modes deviating from normal behaviors based on preset rules; An adaptive multi-agent system is deployed, and each agent is responsible for analyzing different types of SMART attributes; the agents include original error count agents, wear leveling agents and temperature agents; the agents are connected through a distributed consensus mechanism to cooperatively analyze time sequence evolution of the SMART attributes, cross-attribute correlations and multi-modal data fusion with external environment parameters; Multi-dimensional features of the fault modes, historical performance indicators, real-time performance parameters and health performance indicators are extracted; Machine learning models are used to detect subtle performance degradation signs for early warning and generate fault diagnosis reports with time stamps, associated performance and health data; Dynamic error correction code strength verification and data integrity verification under stress are performed; the dynamic error correction code strength verification and data integrity verification method specifically comprises the following steps: A series of adaptive and high-pressure test scenarios are designed to induce bit errors of a solid state disk type flash memory under the conditions of simulating extreme temperature, voltage fluctuation, high write wear and / or specific data mode; the system real-time monitors the error correction capability of the memory, dynamically adjusts at least one of the following operation strategies of the solid state disk type flash memory: flash memory refresh period, write voltage, read reference voltage, bad block management algorithm, ECC algorithm strength switching, read retry mechanism, data refresh and / or data migration strategy, power management strategy and data scrambling strategy, and verifies the data recovery and integrity maintenance capability under extreme conditions; the effectiveness of the internal error correction mechanism of the solid state disk is actively detected and verified; meanwhile, through prediction reliability modeling, the remaining service life is estimated and the fault is predicted by using multi-source heterogeneous data to generate product optimization suggestions for reliability management.

2. The multi-thread based solid state drive reliability automated testing method of claim 1, wherein, The specific steps of extracting the multi-dimensional features include: Features covering time domain, frequency domain and statistical domain are extracted and constructed from the fault modes, historical performance indicators, real-time performance parameters and health performance indicators; the features are screened and reduced in dimension; and the interaction and ratio between the performance indicators and the health indicators are calculated.

3. The multi-thread based solid state drive reliability automated testing method of claim 1, wherein, The machine learning models are used to detect subtle performance degradation signs to realize early warning of the health state of the solid state disk, and specifically include the following steps: Pre-trained machine learning models are used to learn and analyze the multi-dimensional features, detect subtle performance degradation signs within a normal fluctuation range and distinguish the subtle performance degradation signs from incidental noise. When the abnormal score and / or classification result output by the machine learning model reaches a preset early warning threshold, the system triggers an early warning signal and automatically generates a fault diagnosis report with an accurate time stamp and associated specific performance indicators and health data that caused the early warning trigger.

4. The multi-thread based solid state drive reliability automated testing method of claim 1, wherein, The prediction reliability modeling and residual service life estimation method specifically comprises: Integrate multi-source heterogeneous data including continuous performance data of solid state disks, SMART health attributes, manufacturing batch information, historical test results and actual field failure data; use random forest algorithm to build a full life cycle reliability prediction model to estimate the residual service life of solid state disks and predict potential failure time points and failure modes.

5. The multi-thread based solid state drive reliability automated testing method of claim 4, wherein, The product optimization suggestions for reliability management specifically include: Based on the failure time points and the failure modes, analyze the reliability performance of solid state disks under different loads, environments and / or batches, identify the key influencing factors and degradation modes that cause performance degradation and / or potential failure, and summarize the analysis results; According to the analysis results, generate specific product optimization suggestions for solid state disk design, manufacturing process, firmware optimization, test strategy and preventive maintenance plan.

6. A multi-thread based solid state drive reliability automated testing system, characterized in that, Comprise: A high-granularity performance data acquisition module acquires and records real-time performance parameters of solid state disks under multi-thread concurrency with microsecond-level time granularity; combining the real-time performance parameters with deep analysis of SMART information, it captures performance transient fluctuations through multi-thread parallel processing and identifies failure modes; The data integration and preliminary anomaly identification module deeply analyzes the SMART information to obtain internal structure, threshold and correlation between attributes; Configure multiple independent parallel threads to quickly capture minor performance transient fluctuations through real-time cross comparison, and preliminarily identify failure modes that deviate from normal behavior based on preset rules; A multi-dimensional feature engineering module deploys an adaptive multi-agent system, each agent is responsible for analyzing different categories of SMART attributes, including original error count agent, wear leveling agent and temperature agent. These agents analyze the time series evolution of SMART attributes, cross-attribute correlation, and multi-modal data fusion with external environmental parameters through distributed consensus mechanism; extract multi-dimensional features of the failure modes, historical performance indicators, real-time performance parameters and health performance indicators; The multi-dimensional feature engineering module uses machine learning models to detect subtle performance degradation signs for early warning and generate fault diagnosis reports with timestamps, associated performance and health data; The reliability verification and life prediction module is used to perform dynamic error correction code strength verification and data integrity verification under stress, and the dynamic error correction code strength verification and data integrity verification method specifically comprises: A series of adaptive, high-pressure test scenarios are designed to induce bit errors in solid state drive type flash memory under simulated extreme temperature, voltage fluctuation, high write wear and / or specific data pattern conditions; the system monitors the memory's error correction capability in real time, dynamically adjusts at least one of the following operating strategies of the solid state drive type flash memory: flash refresh cycle, write voltage, read reference voltage, bad block management algorithm, ECC algorithm strength switching, read retry mechanism, data refresh and / or data relocation strategy, power management strategy and data scrambling strategy, and verifies the data recovery and integrity maintenance capability under extreme conditions; actively detect and verify the effectiveness of the internal error correction mechanism of the solid state drive; at the same time, through the prediction of reliability modeling, the remaining useful life is estimated and the failure is predicted by using multi-source heterogeneous data, and the product optimization suggestions of reliability management are generated.

Images