Solid state disk storage performance dynamic detection method and system

Abstract

The present application belongs to the technical field of solid state disk detection, and particularly relates to a solid state disk storage performance dynamic detection method and system. The method comprises the following steps: adopting time-sharing background scheduling and multi-dimensional index unified collection to standardize the real business scene for speed measurement; a multi-dimensional feature matching model is constructed to automatically distinguish the root cause of energy decay; long-time load stability is continuously monitored; heat source coupling performance deduction analysis and adaptive control are performed: GPU idle and GPU full load dual working conditions are compared for speed measurement, and software adaptive cache control is performed after determining the heat source + high write dual risk working condition. The present application has the advantages of preventing distortion, accurate positioning and long-term effect.

Description

Technical Field

[0001] This invention relates to the field of solid-state drive (SSD) testing technology, specifically to a method and system for dynamic testing of SSD storage performance. Background Technology

[0002] The description of the background art in this invention pertains to related technologies and is used merely for illustration and to facilitate understanding of the invention. It should not be construed as the applicant explicitly believing or presuming that the invention was prior art on the filing date of the first application.

[0003] In the current era of accelerated digital transformation, solid-state drives (SSDs), with their advantages of low latency, high IOPS, and strong shock resistance, have completely replaced hard disk drives (HDDs) as the mainstream storage medium for personal computing, data center cloud computing, and big data storage. According to industry data, by 2025, the global market share of enterprise-grade SSDs had exceeded 78%. The stability of their storage performance is directly related to the continuity of critical business and data security—business interruptions caused by sudden performance degradation or failure of SSDs can often result in economic losses dozens of times the value of the hardware itself. Dynamic monitoring of SSD storage performance is a key to solving this problem. Common monitoring methods and their limitations are as follows: Existing technology 1: Fixed unloaded single-scenario benchmark speed measurement solution Implementation When the device is idle and has no business load, it only performs a single speed test task of continuous large file read and write, and outputs the theoretical read and write bandwidth under no-load as the basis for SSD performance evaluation. It only collects the single indicator of peak bandwidth and does not distinguish between 4K random and fragmented read and write scenarios.

[0004] Defects exist 1. Simulating only ideal idle conditions cannot reproduce real business scenarios such as daily small file access, long-term fragmentation accumulation, and large file batch copying. The test results deviate greatly from the actual performance, resulting in distorted performance evaluation. 2. It only outputs read and write bandwidth values, without detailed indicators such as IOPS and 99th percentile latency, and cannot reflect hidden performance shortcomings such as random access and high load stuttering; 3. Single instantaneous speed measurement, without 24-hour long-term monitoring, cannot capture continuous speed drop and performance degradation under long-term high load, and cannot assess long-term operational stability; 4. It lacks thermal coupling analysis capabilities, completely ignoring the impact of high-temperature slowdown caused by the heat from the graphics card and CPU, and cannot quantify the performance loss of SSDs due to heat sources.

[0005] Existing technology 2: A general performance speed measurement solution without classification and recognition Implementation After completing the speed test and obtaining the performance degradation result, it simply outputs the conclusion "speed decreased" without any multi-factor feature differentiation logic, making it impossible to locate the root cause of the performance degradation; the speed test consumes a large amount of disk I / O throughout the process, and the speed test will cause system lag when the front-end business is running.

[0006] Defects exist 1. It is impossible to distinguish the four types of performance degradation root causes: high temperature throttling, flash memory aging, file fragmentation, and background process preemption, making it difficult for operation and maintenance personnel to optimize in a targeted manner; 2. Lacking lightweight time-sharing scheduling logic, speed test tasks and user read / write operations compete for disk bandwidth, affecting normal office work and file operations; 3. It does not link graphics card and CPU temperature and power consumption data, and cannot identify the coupling relationship between high temperature of heat source and SSD performance degradation; 4. There is no adaptive caching mechanism. After detecting performance degradation, it only logs the data and cannot actively optimize read and write strategies to alleviate the loss.

[0007] Existing technology 3: Short-duration single-cycle SSD health check solution Implementation It only manually triggers short-term speed tests and has no automatic long-term monitoring mechanism; it only reads the single-point temperature of the SSD controller, does not compare it with the load conditions of surrounding hardware, does not have a heat source temperature difference performance comparison analysis, and does not have the ability to quantify long-term performance degradation.

[0008] Defects exist 1. Relying on manual detection, it cannot automatically and continuously monitor 24-hour time-series performance changes, and cannot capture sudden speed drops under high load and nighttime batch write attenuation; 2. The lack of idle / full load comparison speed tests for the graphics card makes it impossible to quantify the performance degradation caused by direct hot air blowing from the graphics card and to determine the level of heat source interference. 3. Comparing a single speed test to the factory specifications is insufficient to calculate the average performance degradation and failure frequency over long-term use, resulting in a one-sided assessment of lifespan and stability. 4. Without active protection and control measures, it cannot adjust the write strategy after recognizing high-temperature performance degradation, and continuous high-temperature read and write accelerates flash memory aging. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for dynamic testing of solid-state drive (SSD) storage performance. This invention offers advantages such as preventing distortion, accurate positioning, and long-term effectiveness.

[0010] The objective of this invention is achieved through the following technical solutions: On one hand, embodiments of the present invention provide a method for dynamically detecting the storage performance of a solid-state drive, comprising the following steps: Standardized speed testing of real business scenarios is conducted by using time-sharing background scheduling and unified collection of multi-dimensional indicators. Construct a multi-dimensional feature matching model to automatically identify the root causes of new energy decay; Continuous monitoring of long-term load stability; Thermal source coupling performance deduction analysis and adaptive control: The speed test was carried out under the dual conditions of graphics card idle and graphics card full load. After determining the dual risk conditions of thermal source + high write, the software adaptive cache control was implemented.

[0011] Furthermore, the time-sharing background scheduling mechanism is as follows: automatically identify the system IO idle window, start the speed test task in the background in time periods, the speed test process has a lower priority than the user business process, and limit the disk bandwidth usage to within 5%.

[0012] Furthermore, multi-dimensional metrics include: IOPS, average response latency, peak bandwidth, and 99th percentile latency.

[0013] Furthermore, standardized speed tests are conducted in real-world business scenarios, including continuous large file read / write tests, 4K random read / write tests, and fragmented mixed read / write tests.

[0014] Furthermore, continuous large file read / write tests include: simulating video copying, image decompression, and backup archiving scenarios, and collecting peak bandwidth and average latency for continuous read / write. The 4K random read / write test includes: simulating daily scenarios such as system boot, software loading, and small file access in the database, and collecting 4K IOPS and random access latency; The fragmented mixed read / write test includes: simulating long-term disk fragmentation accumulation, scattered file addition and deletion, and mixed random / sequential read / write.

[0015] Furthermore, a multi-dimensional feature matching model is constructed to automatically distinguish the root causes of performance degradation. This model combines SSD native SMART logs, controller temperature, total write / erase volume, disk fragmentation rate, and system process I / O usage data to automatically distinguish four types of performance degradation root causes, including: high temperature throttling, flash memory aging, file fragmentation, and background process preemption.

[0016] Furthermore, the determination of the four types of root causes of performance degradation is carried out using the following method: High-temperature throttling characteristics: When the SSD controller temperature exceeds the 70℃ threshold, read and write bandwidth drop sharply; after the graphics card / CPU is idle, the SSD temperature drops and performance recovers accordingly. Flash memory aging characteristics: The total number of erases and writes across the entire disk is close to the upper limit of the nominal PE lifespan. Read and write latency increases uniformly and synchronously in all continuous and random scenarios, and performance does not recover significantly after cooling. File fragmentation characteristics: Only 4K random read and write performance is significantly reduced, while the read and write speed of consecutive large files does not change significantly, and the disk fragmentation rate index rises in tandem. Background process preemption characteristics: During the speed test, high IO background processes such as system updates, cloud disk synchronization, and backup tools were detected simultaneously. After closing the processes, the performance returned to the baseline level.

[0017] Furthermore, the continuous monitoring of long-term load stability includes: 24-hour time-series data acquisition: Continuously capture real-time disk read / write bandwidth and response latency time-series curves, and sample storage performance by the hour; Stability Quantitative Statistics: Automatically counts the frequency of performance drops and the duration of a single speed drop under high load conditions throughout the day, and calculates the performance stability coefficient; Long-term performance degradation comparison: The difference between real-time multi-scenario speed test indicators and the SSD factory standard benchmark is compared to quantify the overall performance degradation ratio after long-term use of the device.

[0018] Furthermore, the dual-condition comparison speed test involves collecting speed test data of the same SSD under two conditions: graphics card idle and graphics card full load. The performance degradation difference corresponding to the temperature difference of the SSD under the two conditions is calculated to quantify the negative impact of peripheral heat sources such as graphics card hot air and CPU heat on SSD read and write performance. High-temperature coupling risk assessment: When the SSD controller temperature is >68℃ and the disk is continuously subjected to high IO writes, it is determined to be a dual risk condition of heat source + high write. Software adaptive cache control: After identifying dual high-risk scenarios, the operating system kernel automatically increases the disk write cache, splits continuous large block writes into intermittent segmented writes, reduces the duration of continuous high-temperature read and write operations of SSD, and alleviates the performance loss and flash memory aging caused by high temperature.

[0019] On the other hand, embodiments of the present invention provide a dynamic performance testing system for solid-state drives (SSDs), comprising: The multi-scenario lightweight time-sharing performance benchmark test suite module is configured to perform standardized speed testing of real business scenarios using time-sharing background scheduling and unified collection of multi-dimensional indicators. The performance degradation multi-factor classification and identification algorithm module is configured to build a multi-dimensional feature matching model to automatically distinguish the root causes of energy degradation. The long-term load stability continuous monitoring mechanism module is configured for long-term load stability continuous monitoring; The thermal source coupling performance deduction analysis and adaptive control module is configured for thermal source coupling performance deduction analysis and adaptive control: the speed test is compared between the graphics card under idle and full load conditions, and the software adaptive cache control is performed after determining the dual risk conditions of thermal source + high write.

[0020] The embodiments of the present invention have the following beneficial effects: 1. Realistic simulation across multiple scenarios to resolve distortion issues in no-load speed measurement assessment. It covers three mainstream real-world usage scenarios: continuous large files, 4K random, and fragmented mixed data. It collects multi-dimensional read and write metrics, no longer relying solely on theoretical idle speed to judge SSD performance. The test results are highly consistent with users' daily usage experience, accurately pinpointing hidden performance shortcomings such as random stuttering and fragmented speed drops.

[0021] 2. Automatically distinguishes four types of performance degradation causes and accurately locates performance bottlenecks. The multi-factor feature matching algorithm automatically identifies four types of performance degradation causes: high temperature, flash memory aging, file fragmentation, and background preemption. It eliminates the need for manual troubleshooting and quickly provides targeted optimization solutions (cleaning up fragments, closing background processes, improving heat dissipation, and predicting hard drive replacement), significantly reducing the cost of troubleshooting and maintenance.

[0022] 3. Lightweight background automatic monitoring, balancing detection accuracy and business smoothness. It adopts a time-sharing idle scheduling lightweight speed test mechanism, which has low bandwidth consumption and will not cause foreground file copying or software lag; it supports 24 / 7 fully automatic continuous timing monitoring, fully captures continuous speed drop issues under long-term high load, quantifies long-term performance degradation, and achieves long-term tracking of SSD performance.

[0023] 4. Pure software-based thermal source coupling simulation to quantify the performance loss caused by heat generation in peripheral hardware. It eliminates the need for external temperature and humidity sensors, quantifying performance degradation caused by heat sources by comparing the native temperature and load data of the graphics card, CPU, and SSD with the results of dual-condition speed tests. It also provides an adaptive write cache control strategy at the kernel level to proactively reduce the damage to SSD performance and flash memory lifespan caused by high temperature and high IO conditions.

[0024] 5. No hardware modifications required throughout the entire process; compatible with all types of SSD devices. The entire detection, analysis, and control logic is based on the NVMe / SATA standard protocol and the system kernel IO scheduling interface. It requires no external acquisition hardware or modification of the host hardware structure, and is compatible with various NVMe and SATA solid-state drives. It can be deployed universally on desktops, servers, and industrial control equipment.

[0025] 6. Long-term performance health quantification, early prediction of hardware aging failures. By comprehensively evaluating the health status of SSDs from multiple dimensions, including 24-hour timing stability coefficient, long-term performance degradation ratio, and total flash memory write / erase volume, the aging and degradation trend of flash memory can be predicted in advance. Replacement warnings can be pushed before read / write performance drops significantly or bad blocks appear, thus avoiding file copy failures, system lag, and data loss. Detailed Implementation

[0026] The present application will be further described below with reference to the embodiments.

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, in the following description, different "an embodiment" or "an embodiment" do not necessarily refer to the same embodiment. Different embodiments can be substituted or combined, and for those skilled in the art, other implementation methods can be obtained based on these embodiments without creative effort.

[0028] To address the following shortcomings of existing detection methods: fixed benchmark speed tests can only obtain theoretical read and write speeds under idle conditions, failing to reproduce real-world usage scenarios such as fragmented files, large file copies, and system cache writes, resulting in distorted performance evaluations; they cannot distinguish between performance degradation caused by four different factors: hardware aging, high-temperature throttling, file fragmentation, and background synchronization tasks, making it difficult to pinpoint performance bottlenecks; there is no continuous monitoring mechanism for sustained speed drops under prolonged high loads and random 4K read and write degradation, meaning a single speed test cannot reflect long-term stability; and there is no purely software-based thermal coupling performance evaluation method, failing to quantify the negative impact of graphics card heat and CPU heat on SSD read and write performance. This application proposes the following solutions, which will be detailed below with specific embodiments: A method for dynamically testing the storage performance of a solid-state drive (SSD) includes the following steps: Standardized speed testing of real business scenarios is conducted by adopting time-sharing background scheduling and unified collection of multi-dimensional indicators; full simulation of multiple scenarios solves the problem of distortion in idle speed test evaluation: covering three mainstream real-world usage scenarios: continuous large files, 4K random, and fragmented mixed, collecting multi-dimensional read and write indicators, no longer relying solely on the theoretical idle speed to judge SSD performance, the test results are highly consistent with the user's daily usage experience, and accurately locate hidden performance shortcomings such as random stuttering and fragmented speed drop.

[0029] A multi-dimensional feature matching model is constructed to automatically distinguish the root causes of performance degradation. Through a multi-factor feature matching algorithm, four types of performance degradation root causes are automatically identified: high temperature, flash memory aging, file fragmentation, and background preemption. No manual investigation is required, and targeted optimization solutions are quickly provided (cleaning up fragments, closing background processes, improving heat dissipation, and predicting hard drive replacement), which greatly reduces the cost of operation and maintenance fault location.

[0030] Long-term load stability monitoring; by comprehensively evaluating the health status of SSD through multiple dimensions such as 24-hour timing stability coefficient, long-term performance degradation ratio, and total flash memory erase and write volume, the aging and degradation trend of flash memory can be predicted in advance, and replacement warnings can be pushed before read and write performance drops significantly or bad blocks appear, so as to avoid file copy failure, system lag, and data loss.

[0031] Thermal source coupling performance analysis and adaptive control: Dual-condition speed testing with GPU idle and full load identifies the dual-risk conditions of heat source + high write operation, and then implements software adaptive cache control. No external temperature and humidity sensors are required; performance degradation caused by heat sources is quantified by comparing the native temperature and load data of the GPU, CPU, and SSD with the dual-condition speed test results. Simultaneously, a kernel-level adaptive write cache control strategy is provided to proactively reduce the damage to SSD performance and flash memory lifespan caused by high-temperature, high-IO conditions.

[0032] It's important to clarify that the graphics card idle and full-load performance comparison test involves collecting and analyzing core data from the graphics card under both low-load idle and high-load full-load operation. Under idle conditions, the test measures the graphics card's basic idle power consumption, idle temperature, and background silent resource usage, confirming that no abnormal background programs are consuming graphics card resources. Under full load conditions, the test measures the graphics card's performance ceiling and actual frame rate output. By comparing the usage differences between the two conditions, it's easy to determine whether the graphics card itself is underperforming or if the CPU or memory is slowing down the scheduling: for example, if the graphics card is fully loaded during a game but the CPU usage is only 60%, the configuration is reasonable; if the CPU is fully loaded but the graphics card usage is only 60%, the CPU is the bottleneck in the overall system performance, avoiding unnecessary hardware upgrades. Idle conditions confirm that the graphics card's basic operating status is normal, without basic issues like idle screen flickering or abnormally high temperatures. Under full load, a long-term stress test is conducted to check for issues such as aging thermal paste, loose power connections, and memory failures, observing for abnormalities like blue screens, crashes, and screen flickering during full load operation. By comparing the temperature difference between idle and full load conditions, the efficiency of the cooling system can be intuitively assessed. Under idle conditions, the graphics card's standby power consumption can be measured, confirming that its dynamic power-saving technology is functioning correctly. This allows for automatic power reduction and minimizes unnecessary power waste during light scenarios such as daily office work and watching videos. Under full load conditions, the peak actual power consumption of the graphics card can be measured, and by comparing it to the card's nominal TDP / TGP values, it can be determined whether the power consumption meets design standards. By calculating the ratio of the power consumption difference to the performance output, the true energy efficiency ratio of the graphics card can be calculated, providing a direct comparison of the energy consumption performance of different graphics card architectures and helping users find the optimal balance between performance and power consumption. Idle data corresponds to light daily usage scenarios, verifying the smoothness and quietness of desktop operations, web browsing, and 4K video playback. Full load data corresponds to heavy-load scenarios such as AAA games, professional rendering, and AI training, measuring the graphics card's maximum frame rate and rendering speed. By comparing the two scenarios, it can be confirmed that the graphics card will not experience issues such as erratic fan spinning or unexplained stuttering under light scenarios, and that it will not experience temperature-triggered throttling or sudden frame rate drops under high loads. This comprehensively covers the usage needs of users across all scenarios, from daily office work to heavy productivity. Idle load testing alone cannot uncover potential problems under high loads, while full load testing alone cannot distinguish whether abnormally high power consumption originates from the graphics card itself or background programs. By comparing the two conditions, test interference from system background processes and driver anomalies can be eliminated, making the final performance scores, temperature, and power consumption data more reliable and trustworthy than speed tests under a single condition.

[0033] In some embodiments of the present invention, the time-sharing background scheduling mechanism is as follows: It automatically identifies system I / O idle windows, starts speed test tasks in the background at different times, and the priority of the speed test process is lower than that of user business processes, limiting disk bandwidth usage to within 5%. It does not preempt foreground read / write resources; it automatically postpones speed tests during high I / O loads. For example, if a user is compiling a large project locally, the code reading / writing and object file generation of this user process are high I / O tasks. The SMART disk I / O detection process automatically triggered by the background system is set to a high nice value (CPU priority 19) + the lowest I / O priority by default. Even if it continuously initiates disk read detection requests, it will not preempt the compilation process's resources, and the compilation progress will not be slowed down.

[0034] It's important to note that setting the priority of the speed test process below that of user business processes during system speed testing improves the user experience. It avoids preempting system resources used by user processes, ensuring smooth operation of user services (such as app response, interface operation, and file read / write) without lag or delays. The speed test runs silently in the background, virtually imperceptible to users, and does not interfere with normal use. This ensures system stability by preventing the speed test process from consuming excessive core resources, which could lead to insufficient resources for critical business operations, reducing the risk of system lag and crashes. It follows system resource scheduling logic, reducing priority conflicts and resulting in more rational resource allocation and stable operation. It improves the reliability of the speed test results. During idle periods, the system's resource load is low, and by relinquishing high priority, the speed test process more accurately reflects the actual IO / bandwidth capabilities of the system / network under idle conditions. Finally, by avoiding resource contention with user businesses, the speed test results are not skewed downwards due to resource competition, making them more valuable for reference.

[0035] The entire detection, analysis, and control logic is based on the NVMe / SATA standard protocol and the system kernel IO scheduling interface. It requires no external acquisition hardware or modification of the host hardware structure, and is compatible with various NVMe and SATA solid-state drives. It can be deployed universally on desktops, servers, and industrial control equipment.

[0036] In some embodiments of the present invention, the multi-dimensional metrics include: IOPS, average response latency, peak bandwidth, and 99th percentile latency. It should be noted that those skilled in the art can select the metrics according to their needs; the above are preferred embodiments and are not intended to limit the scope of protection of this application.

[0037] In some embodiments of the present invention, standardized speed tests in real business scenarios include continuous large file read / write tests, 4K random read / write tests, and fragmented mixed read / write tests.

[0038] In some embodiments of the present invention, continuous large file read / write tests include: simulating video copying, image decompression, and backup archiving scenarios, and collecting peak bandwidth and average latency of continuous read / write; The 4K random read / write test includes: simulating daily scenarios such as system boot, software loading, and small file access in the database, and collecting 4K IOPS and random access latency; The fragmented mixed read / write test includes: simulating long-term disk fragmentation accumulation, scattered file additions and deletions, and mixed random / sequential read / write operations. The negative impact of fragmentation on performance is quantified.

[0039] In some embodiments of the present invention, a multi-dimensional feature matching model is constructed to automatically distinguish the root causes of performance degradation by combining SSD native SMART logs, controller temperature, total write / erase volume, disk fragmentation rate, and system process I / O usage data. The root causes of performance degradation include: high temperature throttling, flash memory aging, file fragmentation, and background process preemption.

[0040] In some embodiments of the present invention, the determination of the four types of root causes of performance degradation is carried out using the following method: High-temperature throttling characteristics: When the SSD controller temperature exceeds the 70℃ threshold, read and write bandwidth drop sharply; after the graphics card / CPU is idle, the SSD temperature drops and performance recovers accordingly. Flash memory aging characteristics: The total number of erases and writes across the entire disk is close to the upper limit of the nominal PE lifespan. Read and write latency increases uniformly and synchronously in all continuous and random scenarios, and performance does not recover significantly after cooling. File fragmentation characteristics: Only 4K random read and write performance is significantly reduced, while the read and write speed of consecutive large files does not change significantly, and the disk fragmentation rate index rises in tandem. Background process preemption characteristics: During the speed test, high IO background processes such as system updates, cloud disk synchronization, and backup tools were detected simultaneously. After closing the processes, the performance returned to the baseline level.

[0041] In some embodiments of the present invention, the long-term load stability continuous monitoring includes: 24-hour time-series data acquisition: Continuously capture real-time disk read / write bandwidth and response latency time-series curves, and sample storage performance by the hour; For example, the 24-hour disk bandwidth curve: the horizontal axis represents hourly intervals, and the vertical axis represents bandwidth values. It visually shows the trend of bandwidth surging to over 300MB / s during peak business hours (such as 8 AM and 7 PM) and dropping back to below 50MB / s during off-peak hours in the early morning. The 24-hour I / O response latency curve: under normal load, latency is stable in the 1-3ms range. When bandwidth is fully utilized, latency spikes appear, reaching over 15ms. The curve allows for quick identification of specific hourly windows of latency spikes. The sample trend line: connecting hourly aggregated sample points into a smooth curve filters out 5-second-level instantaneous spikes, more clearly showing the overall performance variation of the disk throughout the day, providing data for subsequent storage expansion and load scheduling.

[0042] Stability Quantitative Statistics: Automatically counts the frequency of performance drops and the duration of a single speed drop under high load conditions throughout the day, and calculates the performance stability coefficient; For example, let's take a 1TB enterprise-grade SSD data disk on a Linux cloud server running an e-commerce database as an example to simulate a high-load scenario during an 8-hour promotional event and automatically count the frequency of performance drops: High load judgment: IO utilization consistently ≥80%, consistent with the real scenario of full-load database read / write during a promotional event. Performance drop judgment threshold: Disk random read IOPS deviates from the baseline value by ±5%, and await (average IO wait time) exceeds 20ms. Only when both conditions are met is it considered a valid drop, avoiding misjudging normal small fluctuations. Monitoring sampling: The system uses the iostat tool to automatically collect data at 1-minute granularity, generating a total of 480 statistical samples throughout the day. The system background automatically traverses all 480 sets of sampled data throughout the day, filtering out events that simultaneously trigger both drop thresholds, ultimately counting a total of 6 valid performance drops throughout the day, each corresponding to a real business scenario. The system automatically generates a statistical report: Under high load conditions throughout the day, the disk experienced a total of 6 performance drops. It also automatically associates the process ID and business operation type corresponding to each drop and outputs it directly to the operation and maintenance personnel for subsequent optimization reference, without requiring manual line-by-line log checking.

[0043] Long-term performance degradation comparison: The difference between real-time multi-scenario speed test indicators and the SSD factory standard benchmark is compared to quantify the overall performance degradation ratio after long-term use of the device.

[0044] In some embodiments of the present invention, the dual-condition comparison speed test is as follows: the speed test data of the same SSD under two conditions, namely, the graphics card is idle and the graphics card is fully loaded, respectively, and the performance degradation difference corresponding to the temperature difference of the SSD under the two conditions is calculated to quantify the negative impact of peripheral heat sources such as graphics card hot air and CPU heat on SSD read and write performance. High-temperature coupling risk assessment: When the SSD controller temperature is >68℃ and the disk is continuously subjected to high IO writes, it is determined to be a dual risk condition of heat source + high write. Software adaptive cache control: After identifying dual high-risk scenarios, the operating system kernel automatically increases the disk write cache, splits continuous large block writes into intermittent segmented writes, reduces the duration of continuous high-temperature read and write operations of SSD, and alleviates the performance loss and flash memory aging caused by high temperature.

[0045] On the other hand, embodiments of the present invention provide a dynamic performance testing system for solid-state drives (SSDs), comprising: The multi-scenario lightweight time-sharing performance benchmark test suite module is configured to perform standardized speed testing of real business scenarios using time-sharing background scheduling and unified collection of multi-dimensional indicators. The performance degradation multi-factor classification and identification algorithm module is configured to build a multi-dimensional feature matching model to automatically distinguish the root causes of energy degradation. The long-term load stability continuous monitoring mechanism module is configured for long-term load stability continuous monitoring; The thermal source coupling performance deduction analysis and adaptive control module is configured for thermal source coupling performance deduction analysis and adaptive control: the speed test is compared between the graphics card under idle and full load conditions, and the software adaptive cache control is performed after determining the dual risk conditions of thermal source + high write.

[0046] It should be noted that the above embodiments can be freely combined as needed. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamically testing the storage performance of a solid-state drive (SSD), characterized in that, Includes the following steps: Standardized speed testing of real business scenarios is conducted by using time-sharing background scheduling and unified collection of multi-dimensional indicators. Construct a multi-dimensional feature matching model to automatically identify the root causes of new energy decay; Continuous monitoring of long-term load stability; Thermal source coupling performance deduction analysis and adaptive control: The speed test was carried out under the dual conditions of graphics card idle and graphics card full load. After determining the dual risk conditions of thermal source + high write, the software adaptive cache control was implemented.

2. The method for dynamic testing of solid-state drive storage performance according to claim 1, characterized in that, The time-sharing background scheduling mechanism is as follows: it automatically identifies the system's I / O idle window, starts the speed test task in the background in different time periods, the speed test process has a lower priority than the user's business process, and limits the disk bandwidth usage to within 5%.

3. The method for dynamic testing of solid-state drive storage performance according to claim 1, characterized in that, Multi-dimensional metrics include: IOPS, average response latency, peak bandwidth, and 99th percentile latency.

4. The method for dynamic testing of solid-state drive storage performance according to claim 1, characterized in that, Standardized speed tests in real-world business scenarios include continuous large file read / write tests, 4K random read / write tests, and fragmented mixed read / write tests.

5. The method for dynamic testing of solid-state drive storage performance according to claim 4, characterized in that, The continuous large file read / write test includes: simulating video copying, image decompression, and backup archiving scenarios, and collecting the peak bandwidth and average latency of continuous read / write. The 4K random read / write test includes: simulating daily scenarios such as system boot, software loading, and small file access in the database, and collecting 4K IOPS and random access latency; The fragmented mixed read / write test includes: simulating long-term disk fragmentation accumulation, scattered file addition and deletion, and mixed random / sequential read / write.

6. The method for dynamic testing of solid-state drive storage performance according to claim 1, characterized in that, A multi-dimensional feature matching model is constructed to automatically distinguish the root causes of performance degradation. This model combines SSD native SMART logs, controller temperature, total write / erase volume, disk fragmentation rate, and system process I / O usage data to automatically distinguish four types of performance degradation root causes, including: high temperature throttling, flash memory aging, file fragmentation, and background process preemption.

7. The method for dynamic testing of solid-state drive storage performance according to claim 6, characterized in that, The following method is used to determine the four types of root causes of performance degradation: High-temperature throttling characteristics: When the SSD controller temperature exceeds the 70℃ threshold, read and write bandwidth drop sharply; after the graphics card / CPU is idle, the SSD temperature drops and performance recovers accordingly. Flash memory aging characteristics: The total number of erases and writes across the entire disk is close to the upper limit of the nominal PE lifespan. Read and write latency increases uniformly and synchronously in all continuous and random scenarios, and performance does not recover significantly after cooling. File fragmentation characteristics: Only 4K random read and write performance is significantly reduced, while the read and write speed of consecutive large files does not change significantly, and the disk fragmentation rate index rises in tandem. Background process preemption characteristics: During the speed test, high IO background processes such as system updates, cloud disk synchronization, and backup tools were detected simultaneously. After closing the processes, the performance returned to the baseline level.

8. The method for dynamic testing of solid-state drive storage performance according to claim 1, characterized in that, The aforementioned long-term load stability continuous monitoring includes: 24-hour time-series data acquisition: Continuously capture real-time disk read / write bandwidth and response latency time-series curves, and sample storage performance by the hour; Stability Quantitative Statistics: Automatically counts the frequency of performance drops and the duration of a single speed drop under high load conditions throughout the day, and calculates the performance stability coefficient; Long-term performance degradation comparison: The difference between real-time multi-scenario speed test indicators and the SSD factory standard benchmark is compared to quantify the overall performance degradation ratio after long-term use of the device.

9. The method for dynamic testing of solid-state drive storage performance according to claim 1, characterized in that, Dual-condition comparison speed test: Collect speed test data of SSD of the same specification under two conditions: graphics card idle and graphics card full load. Calculate the performance degradation difference corresponding to the temperature difference of SSD under the two conditions to quantify the negative impact of peripheral heat sources such as graphics card hot air and CPU heat on SSD read and write performance. High-temperature coupling risk assessment: When the SSD controller temperature is >68℃ and the disk is continuously writing at high I / O, it is determined to be a dual risk condition of heat source + high write. Software adaptive cache control: After identifying dual high-risk scenarios, the operating system kernel automatically increases the disk write cache, splits continuous large block writes into intermittent segmented writes, reduces the duration of continuous high-temperature read and write operations of SSD, and alleviates the performance loss and flash memory aging caused by high temperature.

10. A dynamic performance testing system for solid-state drives (SSDs), characterized in that, include: The multi-scenario lightweight time-sharing performance benchmark test suite module is configured to perform standardized speed testing of real business scenarios using time-sharing background scheduling and unified collection of multi-dimensional indicators. The performance degradation multi-factor classification and identification algorithm module is configured to build a multi-dimensional feature matching model to automatically distinguish the root causes of energy degradation. The long-term load stability continuous monitoring mechanism module is configured for long-term load stability continuous monitoring; The thermal source coupling performance deduction analysis and adaptive control module is configured for thermal source coupling performance deduction analysis and adaptive control: the speed test is compared between the graphics card under idle and full load conditions, and the software adaptive cache control is performed after determining the dual risk conditions of thermal source + high write.