Method and system for establishing hard disk performance problem classification model, and analysis method

Abstract

The application provides a hard disk performance problem classification model establishing method and system, and an analysis method. The hard disk performance problem classification model establishing method comprises using an analysis device to perform the following steps: obtaining multiple pieces of training data of multiple single hard disks, each piece of training data comprising multiple vibration parameters and having a preset output result; inputting the multiple pieces of training data into a class neural network model; training the class neural network so that the class neural network can output corresponding multiple preset output results according to the multiple vibration parameters of the multiple pieces of training data; and taking the trained class neural network as a hard disk performance problem classification model. The hard disk performance problem classification model established by the foregoing method successfully finds out the cause of the reduction of the hard disk performance.

Description

Technical Field

[0001] This invention relates to a method for establishing a classification model, and more particularly to a method for establishing a classification model for hard disk performance issues. Background Technology

[0002] With the development of the Internet, the amount of information processed is increasing, and the number of servers required is also increasing. If the server performance decreases, it will affect the data transmission of the entire network, which may cause game crashes, email transmission failures, or video conference interruptions. How to improve server performance has become an important issue.

[0003] Generally speaking, server performance is related to hard drive performance; server performance is affected by hard drive performance. If hard drive performance declines, server performance will also decline. Engineers often use manual, step-by-step analysis to find the cause of declining hard drive performance within the server, but frequently fail to find the root cause affecting hard drive performance, making it impossible to address the problem effectively. Summary of the Invention

[0004] Based on the foregoing, the present invention provides a method and system for establishing a classification model of hard disk performance problems, as well as an analysis method, in order to find the root cause of hard disk performance issues.

[0005] A method for establishing a hard disk performance problem classification model according to an embodiment of the present invention includes, performed by an analysis device: acquiring multiple training data corresponding to multiple individual hard disks, each training data including multiple vibration parameters and having multiple preset output results, each preset output result indicating one of the multiple performance problems; inputting the training data into a neural network model, calculating multiple first output results corresponding to the training data, wherein the neural network model has weight reconfiguration; performing a weight adjustment procedure based on the difference between the first output results and the preset output results, the weight adjustment procedure including: adjusting the weight reconfiguration, and generating multiple second output results using the neural network model based on the adjusted weight reconfiguration and the training data; if the second output results do not correspond to the preset output results, performing a weight adjustment procedure based on the difference between the second output results and the preset output results; if the second output results correspond to the preset output results, using the neural network model with the adjusted weight reconfiguration as the hard disk performance problem classification model.

[0006] A hard disk performance problem analysis method according to an embodiment of the present invention includes the following steps performed by a computer system: obtaining the aforementioned hard disk performance problem classification model; inputting measurement data of an abnormal server hard disk into the hard disk performance problem classification model to generate a classification result, wherein the classification result is one of a plurality of performance problems.

[0007] A hard disk performance problem classification model establishment system according to an embodiment of the present invention includes multiple vibration parameter measuring devices, an input device, and an analysis device. The multiple vibration parameter measuring devices are used to measure multiple vibration parameters of each of multiple individual hard disks. The input device is used to receive a preset output result for each individual hard disk, the preset output result being one of multiple performance problems. The analysis device is connected to these vibration parameter measuring devices and the input device, and includes a neural network model. The analysis device performs the following steps: acquiring multiple training data corresponding to these individual hard disks, each training data including the vibration parameters corresponding to these individual hard disks and possessing the preset output results; performing a weight adjustment procedure based on the differences between these first output results and these preset output results, the weight adjustment procedure including: adjusting the weight reassembly, and generating multiple second output results using the neural network model based on the adjusted weight reassembly and the training data; if these second output results do not correspond to these preset output results, performing the weight adjustment procedure based on the differences between these second output results and these preset output results; if these second output results correspond to these preset output results, using the neural network model with the adjusted weight reassembly as the hard disk performance problem classification model.

[0008] In summary, the method and system for establishing a hard disk performance problem classification model of the present invention are based on a neural network model. The neural network model is trained using training data of multiple vibration parameters associated with a single hard disk. The weights of the neural network are adjusted based on the differences between the preset output results and the output results of the neural network model based on the multiple training data, thereby establishing a hard disk performance problem classification model with high accuracy. Furthermore, the hard disk performance problem analysis method of the present invention, by inputting measurement data of faulty hard disks within the service storage system into the aforementioned hard disk performance problem classification model, can effectively deduce the main causes of reduced hard disk performance. Attached Figure Description

[0009] Figure 1 This is a functional block diagram of a system for establishing a classification model of hard disk performance problems, based on an embodiment of the present invention.

[0010] Figure 2 This is a flowchart illustrating a method for establishing a classification model of hard disk performance problems according to an embodiment of the present invention.

[0011] Figure 3 This is a schematic diagram of a neural network-like model illustrated according to an embodiment of the present invention.

[0012] Figure 4 This is a schematic diagram illustrating the execution environment of a hard disk performance problem analysis method according to an embodiment of the present invention.

[0013] Figure 5 This is a flowchart illustrating a method for analyzing hard disk performance problems according to an embodiment of the present invention.

[0014] Component designation explanation

[0015] 1. System for Establishing a Classification Model for Hard Drive Performance Issues

[0016] 2. Computer System

[0017] 11 Vibration Parameter Measuring Instrument

[0018] 12 Analytical apparatus

[0019] 111 Triaxial Accelerometer

[0020] 112 Sound pressure meter

[0021] 113 Resonance Frequency Analyzer

[0022] HL hidden layer

[0023] OUT output layer

[0024] Steps S11 to S17

[0025] Steps S21 to S22

[0026] x1~x4 input points

[0027] Output points y1~y4 Detailed Implementation

[0028] The following detailed description of the features and advantages of the present invention in the embodiments is sufficient to enable anyone skilled in the art to understand the technical content of the present invention and implement it accordingly. Based on the disclosure, patent claims, and drawings in this specification, anyone skilled in the art can easily understand the related objectives and advantages of the present invention. The following embodiments further illustrate the points of the present invention in detail, but are not intended to limit the scope of the present invention in any way.

[0029] It should be understood that although the terms "first," "second," etc., may be used in this invention to describe various components, parts, regions, layers, and / or portions, these components, parts, regions, layers, and / or portions should not be limited by these terms. These terms are used only to distinguish one component, part, region, layer, and / or portion from another component, part, region, layer, and / or portion.

[0030] Additionally, the terms "comprising" and / or "including" refer to the presence of the stated features, areas, wholes, steps, operations, components, and / or parts, but do not exclude the presence or addition of one or more other features, areas, wholes, steps, operations, components, parts, and / or combinations thereof.

[0031] Please see Figure 1 This is a functional block diagram of a system for establishing a hard disk performance problem classification model according to an embodiment of the present invention. Figure 1 As shown, the hard disk performance problem classification model establishment system 1 includes multiple vibration parameter measuring devices 11, an analysis device 12, and an input device 13, wherein the analysis device 12 and the input device 13 are connected to the multiple vibration parameter measuring devices 11.

[0032] The term "single hard drive" as used below refers to an individual hard drive device that can be installed in a server. A server can install multiple hard drive devices, and depending on its expandability, it can install 4, 8, or more than 12 hard drive devices. Multiple vibration parameter measuring devices 11 are used to measure multiple values ​​of multiple vibration parameters for each of the multiple single hard drives. More specifically, the multiple vibration parameter measuring devices 11 may include at least two of a triaxial accelerometer 111, a sound pressure level meter 112, and a resonant frequency analyzer 113. The triaxial accelerometer 111 is used to measure the acceleration and angular acceleration values ​​of the multiple single hard drives. The sound pressure level meter 112 is used to measure the sound pressure level of the multiple single hard drives. The resonant frequency analyzer 113, which can be implemented using a hard drive I / O performance benchmarking tool (IOMeter) or a spectrum analyzer, is used to analyze the resonant frequency values ​​of the multiple single hard drives. That is, the multiple vibration parameters may include at least two of acceleration, angular acceleration, sound pressure level, and resonant frequency.

[0033] Input device 13 can be a user input interface, such as a keyboard and mouse set or a touch display panel; the foregoing is merely an example and not limited to the scope listed in this invention. Input device 13 is used to receive the preset output result of each individual hard drive. The preset output result is one of multiple performance problems, such as acceleration, rotation, sound pressure level, and frequency. In other words, the user can input the preset output result of each individual hard drive to the analysis device 12 through input device 13. Furthermore, the multiple performance problems correspond to the aforementioned multiple vibration parameters. For example, performance problems may include acceleration, rotation, sound pressure level, and frequency, corresponding to acceleration, angular acceleration, sound pressure level, and resonant frequency, respectively. The preset output result can also be determined by an external processor or the user based on the failure condition corresponding to each vibration parameter, and then input to the analysis device 12 through input device 13. Furthermore, if the measured value of a certain vibration parameter meets the failure condition, the preset output result will be set as the performance problem corresponding to this vibration parameter. For example, if the acceleration value meets the failure condition, the default output will be set to an acceleration problem.

[0034] For example, the failure condition settings for a 1TB or 2TB hard drive are: frequency 300Hz / 900Hz, sound pressure level greater than 114dB, and angular acceleration greater than 9rad / s². For a 10TB hard drive, the failure conditions are: frequency 300Hz / 900Hz, sound pressure level greater than 108dB, acceleration greater than 0.2m / s², and angular acceleration greater than 6rad / s². The frequencies of the failure conditions specifically refer to the resonant frequencies of the server chassis and fans.

[0035] The analysis device 12 can be a microcontroller, graphics processor, or other electronic device with data processing and storage functions, and is not limited to the scope listed in this invention. The analysis device 12 receives multiple values ​​of multiple vibration parameters of each individual hard disk and preset output results, and uses them as input to a neural network model to train the neural network. The analysis device 12 uses the trained neural network model as a classification model for hard disk performance problems. The detailed training process of the neural network model will be described later.

[0036] In one embodiment, multiple vibration parameter measuring devices 11, analysis devices 12, and input devices 13 are integrated into a single electronic device. In another embodiment, the analysis device 12 and input devices 13 are integrated into a single electronic device, independently configured with respect to the multiple vibration parameter measuring devices 11, and electrically or communicatively connected to the multiple vibration parameter measuring devices 11. In yet another embodiment, the multiple vibration parameter measuring devices 11, analysis devices 12, and input devices 13 are independently configured, and the analysis device 12 can be located at an edge or in the cloud and communicatively connected to the multiple vibration parameter measuring devices 11 and input devices 13.

[0037] Please refer to the following: Figure 1 and 2 ,in Figure 2 This is a flowchart illustrating a method for establishing a classification model for hard disk performance problems according to an embodiment of the present invention. Figure 2 As shown, the method for establishing a classification model for hard disk performance problems includes steps S11 to S17. Figure 2 The proposed classification model for hard drive performance issues can be applied to... Figure 1 The hard drive performance problem classification model shown in System 1 is used to establish the system, but it is not limited to this. The following examples illustrate this. Figure 1 The operation of the hard disk performance problem classification model establishment system 1 is shown to illustrate steps S11 to S17.

[0038] Step S11: The analysis device 12 acquires multiple training data points corresponding to multiple individual hard drives. Each training data point includes multiple vibration parameters and has multiple preset output results, each preset output result indicating one of multiple performance issues. As mentioned above, the analysis device 12 can acquire multiple values ​​of multiple vibration parameters, such as acceleration, angular acceleration, sound pressure level, and resonant frequency, obtained from multiple vibration parameter measuring devices 11 by measuring each individual hard drive. The analysis device 12 can receive the preset output results of each individual hard drive from the input device 13. In one embodiment, the analysis device 12 can control multiple vibration parameter measuring devices 11 to measure each individual hard drive and transmit the measurement results back. In another embodiment, the multiple vibration parameter measuring devices 11 can be controlled by a user or other control device to measure each individual hard drive and then transmit the measurement results to the analysis device 12. In addition, in step S11, the analysis device 12 can also control multiple vibration parameter measuring devices 11 to measure the server's chassis or fan to obtain multiple values ​​of multiple vibration parameters, such as the resonant frequency of the server's chassis or fan, which can also be used as training data.

[0039] Step S12: Input the multiple training data points into the neural network model using the analysis device 12, and calculate the multiple first output results corresponding to the multiple training data points, wherein the neural network model has weighted recombination. Specifically, as shown in... Figure 3 As shown, the neural network-like model includes an input layer IN, a hidden layer HL, and an output layer OUT. The weighted array contains the weights of the connections between neurons. Specifically, the activation function of the neural network-like model is a normalized function (Softmax), which is a continuous function after differentiation. When the input value is between 0 and 1, the risk of non-activation is greatly reduced, and the misclassification rate is also significantly reduced. The analysis device 12 inputs each piece of training data into the input layer IN. Each piece of training data passes through the input layer IN and the hidden layer HL, and outputs the first output result from the output layer OUT. The input layer IN includes four input points x1 to x4, and the output layer OUT has four output points y1 to y4. For example, input points x1 to x4 represent acceleration, angular acceleration, sound pressure level, and resonant frequency, respectively, while output points y1 to y4 represent acceleration, rotation, sound pressure level, and frequency problems, respectively. The acceleration, angular acceleration, sound pressure level, and resonant frequency of each training data point are input from input points x1 to x4. The neural network model calculates the shortest path for each training data point and generates a corresponding first output result. This first output result indicates that one of the four output points y1 to y4 is the endpoint of the shortest path, with the output value of that endpoint being 1, while the output values ​​of the other output points are 0. It is important to note that... Figure 3 The number of input and output points is illustrated only by way of example and is not intended to limit the invention.

[0040] Step S13: The analysis device 12 adjusts the weight regroup based on the differences between the plurality of first output results and the plurality of preset output results. Specifically, the analysis device 12 calculates the differences between the plurality of first output results and the plurality of preset output results using a cost function, uses this difference as the basis for adjusting the weight regroup, and adjusts the weight regroup using a backpropagation algorithm and stochastic gradient descent. Further, the cost function is cross-entropy. When the differences are too large, the cross-entropy penalty on the neural network model will be greater, improving the accuracy of the neural network model during training.

[0041] Step S14: The analysis device 12 uses a neural network model to generate multiple second output results based on the adjusted weight reassembly and the multiple training data. Specifically, after step S13, the weight reassembly of the neural network model has been adjusted. The analysis device 12 then inputs the training data back into the weighted and adjusted neural network model to generate a new batch of output results (second output results).

[0042] Step S15: The analysis device 12 determines whether the plurality of second output results correspond to the plurality of preset output results. Specifically, the analysis device 12 uses a cost function to calculate the difference between the plurality of second output results and the plurality of preset output results. The cost function, as previously described, can be cross-entropy. If the difference exceeds a preset threshold, it indicates that the plurality of second output results do not correspond to the plurality of preset output results, and the analysis device 12 executes step S16; if the difference does not exceed the preset threshold, it indicates that the plurality of second output results correspond to the plurality of preset output results, and the analysis device 12 executes step S17. The actual value of the preset threshold can be set according to actual needs, and this invention does not limit it.

[0043] Step S16: The analysis device 12 adjusts the weight reassembly based on the differences between these second output results and these preset output results. Specifically, the analysis device 12 uses the differences between the multiple second output results calculated using the cost function and the multiple preset output results as the basis for adjusting the weight group, and uses the backpropagation algorithm and stochastic gradient descent method to adjust the weight reassembly, and then executes steps S14 and S15 again. In other words, the analysis device 12 will repeatedly execute steps S14 to S16 to adjust the weight reassembly of the neural network model until the differences calculated using the cost function do not exceed a preset threshold. In particular, the above-mentioned steps of adjusting the weight reassembly and generating multiple second output results based on the adjusted weight reassembly and training data by the neural network model can be regarded as a weight adjustment procedure, and the multiple second output results and the adjusted weight reassembly generated in each round of the weight adjustment procedure are different.

[0044] Step S17: The analysis device 12 uses the adjusted weighted reconfiguration neural network model as a classification model for hard disk performance problems. If the multiple second output results correspond to the multiple preset output results, it indicates that the adjusted weighted reconfiguration neural network model has been trained. Therefore, in this step, the analysis device 12 uses the trained neural network model as a classification model for hard disk performance problems, which can be used to classify hard disk performance problems on new measurement data.

[0045] Please see Figure 4 and Figure 5 This is a schematic diagram of the execution environment of a hard disk performance problem analysis method according to an embodiment of the present invention, and a flowchart of the hard disk performance problem analysis method according to an embodiment of the present invention. Figure 4 and Figure 5 As shown, the execution environment corresponding to the hard disk performance problem analysis method of the present invention includes a hard disk performance problem classification model establishment system 1 and a computer system 2. The hard disk performance problem classification model establishment system 1 is as follows: Figure 1 The hard disk performance problem classification model shown has been described in detail in the preceding paragraphs and will not be repeated here. Computer system 2 includes a processor that inputs measurement data of the abnormal server hard disk into the hard disk performance problem classification model to obtain classification results. The processor may be, for example, a microcontroller, a graphics processor, or other electronic device with data processing and storage capabilities, and is not limited to the scope set forth in this invention. In this embodiment, the computer system 2 executing the hard disk performance problem classification model and the analysis device establishing the hard disk performance problem classification model are different devices. In another embodiment, the computer system 2 executing the hard disk performance problem classification model and the analysis device establishing the hard disk performance problem classification model are the same device.

[0046] Therefore, the hard disk performance problem analysis method of the present invention includes steps S21 to S22. Step S21: Obtain as follows Figure 2 The steps shown establish a hard disk performance problem classification model. Specifically, the hard disk performance problem classification model can be implemented by a software program. In one embodiment, the software program is stored on a server or external hard disk, which can be connected to the processor of computer system 2. The server transmits the hard disk performance problem classification model to the processor of computer system 2 via a network, or the external hard disk transmits the hard disk performance problem classification model to the processor via a transmission line connected to computer system 2. In other words, the processor of computer system 2 can access and execute the software program of the hard disk performance problem classification model from the server or external hard disk. In another embodiment, the software program of the hard disk performance problem classification model is stored in the storage device of computer system 2, and the processor of computer system 2 reads and executes the software program of the hard disk performance problem classification model.

[0047] Step S22: Computer system 2 inputs the measurement data of the abnormal server hard drive into the hard drive performance problem classification model to obtain the classification result. The classification result indicates one of several performance problems, and these multiple performance problems are associated with multiple vibration parameters. For example, multiple performance problems may include acceleration problems, rotational problems, sound pressure level problems, and frequency problems, which are respectively associated with acceleration, angular acceleration, sound pressure level, and resonant frequency. Specifically, computer system 2 can use the hard drive performance problem classification model to determine the most severe performance problem affecting the server hard drive and identify it as the main cause of the server hard drive problem.

[0048] In one embodiment of the present invention, the method and system for establishing the hard disk performance problem classification model and the analysis method of the present invention can analyze and test the hard disk installed on the server to improve the reliability of the server, making the server suitable for artificial intelligence (AI) computing, edge computing, and can also be used as a 5G server, cloud server or vehicle networking server.

[0049] In summary, the method and system for establishing a hard disk performance problem classification model of the present invention are based on a neural network model. The neural network model is trained using training data of multiple vibration parameters associated with a single hard disk. The weights of the neural network are adjusted based on the differences between the preset output results and the output results of the neural network model based on the multiple training data, thereby establishing a hard disk performance problem classification model with high accuracy. Furthermore, the hard disk performance problem analysis method of the present invention, by inputting measurement data of faulty hard disks within the service storage system into the aforementioned hard disk performance problem classification model, can effectively deduce the main causes of reduced hard disk performance.

[0050] While the present invention has been disclosed above with reference to the foregoing embodiments, it is not intended to limit the invention. Any modifications and refinements made without departing from the spirit and scope of the invention are within the scope of patent protection of the present invention. For details regarding the scope of protection defined in the present invention, please refer to the appended claims.

Claims

1. A method for establishing a classification model for hard disk performance problems, comprising using an analysis device to perform: Multiple training data points corresponding to multiple individual hard drives are obtained. Each training data point includes multiple vibration parameters and has multiple preset output results. Each preset output result indicates one of multiple performance problems. The vibration parameters include multiple of acceleration, angular acceleration, sound pressure level, and resonant frequency. The performance problems include acceleration problems, rotation problems, sound pressure level problems, or frequency problems. The training data is input into a type of neural network model, and multiple first output results corresponding to the training data are calculated, wherein the type of neural network model has a weighted recombination. Based on the difference between the first output result and the preset output result, a weight adjustment procedure is executed, which includes: The weighting is adjusted, and the neural network model of this type generates multiple second output results based on the adjusted weighting and the training data. If the second output result does not correspond to the preset output result, the weight adjustment procedure is executed based on the difference between the second output result and the preset output result; and If the second output result corresponds to the preset output result, the neural network model with the adjusted weight group will be used as a hard disk performance problem classification model. The difference between the first output result and the preset output result is generated by a cost function, and the difference between the second output result and the preset output result is generated by the cost function, which is cross-entropy.

2. The method for establishing a classification model for hard disk performance problems as described in claim 1, wherein adjusting the weighting system is performed using a backpropagation algorithm and a stochastic gradient descent method.

3. The method for establishing a classification model for hard disk performance problems as described in claim 1, wherein the activation function of such neural network model is a normalization function.

4. A method for analyzing hard disk performance problems, comprising using a computer system to perform: Obtain the hard disk performance problem classification model established by the method for establishing a hard disk performance problem classification model as described in any one of claims 1 to 3; and Input measurement data of an abnormal server hard drive into a hard drive performance problem classification model to produce a classification result, wherein the classification result indicates one of the performance problems; the measurement data includes multiple vibration parameters.

5. A system for establishing a classification model for hard disk performance problems, comprising: Multiple vibration parameter measuring instruments are used to measure multiple vibration parameters of each of multiple individual hard drives; An input device is configured to receive multiple preset output results corresponding to the single hard disk, each preset output result indicating one of multiple performance problems; the vibration parameters include multiple of acceleration, angular acceleration, sound pressure level, and resonant frequency; the performance problems include acceleration problems, rotational problems, sound pressure level problems, or frequency problems; and An analysis device, connected to the vibration parameter measuring instrument and the input device, and including a neural network model, performs the following steps: Obtain multiple training data points corresponding to the individual hard drives, each training data point including the vibration parameters of the corresponding individual hard drive and having the preset output result; The training data is input into a type of neural network model, and multiple first output results corresponding to the training data are calculated, wherein the type of neural network model has a weighted recombination. Based on the difference between the first output result and the preset output result, a weight adjustment procedure is executed, which includes: The weighting is adjusted, and the neural network model of this type generates multiple second output results based on the adjusted weighting and the training data. If the second output result does not correspond to the preset output result, the weight adjustment procedure is executed based on the difference between the second output result and the preset output result; and If the second output result corresponds to the preset output result, the neural network model with the adjusted weights will be used as a hard disk performance problem classification model. The difference between the first output result and the preset output result is generated by a cost function, and the difference between the second output result and the preset output result is generated by the cost function, which is cross-entropy.

6. The system for establishing a classification model for hard disk performance problems as described in claim 5, wherein adjusting the weighting system is performed using a backpropagation algorithm and a stochastic gradient descent method.

Images